Dairy Science & Technology · Dairy Processing & Product Technology · also Judging of Dairy Products, Sensory Science
Sensory science applied to dairy: panels, test methods, the official scorecards, and product defects by sight, smell and taste.
In the exams: ICAR AIEEA-PG group 17 Unit I (sensory evaluation and judging of milk and milk products); ICAR AICE subject 35a Unit 3 (judging and grading of milk products); ACS Certified Cheese Professional (assessment and evaluation)
Consumers judge dairy products with their senses before anything else. A carton of milk with a faint cardboard taste, a butter that is greasy and leaky, an ice cream that feels icy, or a cheese that is bitter will be rejected however safe and nutritious it is. Dairy science has therefore always had a strong tradition of sensory judging: experienced graders scored butter and cheese for payment and export long before laboratory instruments existed, and student dairy-products judging contests have been held for over a century. Modern sensory science adds to this tradition the methods of experimental psychology and statistics, so that sensory results can be as reliable as chemical ones.
This manual covers sensory evaluation as applied to dairy products. It begins with the physiology of taste, smell, touch, sight and hearing and the psychological errors that affect judgement. It describes how panels are selected, screened and trained; how discrimination tests such as the triangle, duo-trio and paired comparison work; and how descriptive and hedonic tests measure what products are like and how much people like them. It explains the structure and use of scorecards for milk, butter, ghee, cheese, ice cream and fermented milks, and catalogues the flavour, body and texture defects of dairy products with their causes. It ends with the statistics of sensory data, the links between instrumental and sensory measurement, and consumer testing and sensory end points for shelf life.
Postgraduate entrance examinations in dairy technology ask about sensory evaluation and the judging of milk and milk products, doctoral fellowship examinations about judging and grading, and professional cheese certifications about assessment and evaluation. Typical questions concern the chance probabilities of discrimination tests, the causes of specific defects, panel procedures and scorecard structure.
Taste, or gustation, is the sense that detects non-volatile substances dissolved in saliva. Taste receptor cells are grouped in taste buds, which sit mainly on three kinds of papillae on the tongue: fungiform papillae on the front and sides, foliate papillae on the rear edges and circumvallate papillae in a V across the back; the most numerous papillae, the filiform, carry no taste buds and sense touch. Taste buds also occur on the soft palate and throat. The popular 'tongue map', in which each taste is sensed on one region, is a misreading of early research; all taste qualities can be detected wherever taste buds are found, with only small differences in sensitivity.
Five basic tastes are widely accepted: sweet, sour, salty, bitter and umami, the savoury taste of glutamate and certain nucleotides. Sweet, bitter and umami are detected by G-protein-coupled receptors; sour and salty mainly by ion channels. A taste for fatty acids and the sensation called kokumi, a mouthfulness enhanced by certain peptides, are proposed additional qualities. In dairy products, sweetness comes from lactose and added sugars, sourness from lactic acid in fermented products, saltiness from added salt in butter and cheese and from chloride in mastitic milk, bitterness from peptides produced by proteolysis, and umami from free glutamate in ripened cheeses.
Taste intensity depends on concentration, temperature, the other components present and the individual. Sensitivity to bitter compounds varies genetically between people, and taste acuity declines with age and with smoking.
Smell, or olfaction, detects volatile compounds. Odour molecules reach the olfactory epithelium high in the nasal cavity in two ways: orthonasally, by sniffing through the nostrils, and retronasally, from the mouth to the back of the nose during eating and swallowing. Humans have several hundred types of functional olfactory receptor, and each odorant activates a pattern of them, so that thousands of odours can be distinguished. Most of what people call taste in everyday speech, the difference between strawberry and vanilla, or between fresh and oxidised milk, is in fact retronasal smell, which is why food tastes flat when the nose is blocked.
Flavour is the combined perception of taste, smell and the chemical sensations of the trigeminal nerve, called chemesthesis, which detects pungency, cooling, astringency and the tingle of carbon dioxide. Dairy flavours are subtle and are often defined by compounds present at extremely low concentrations: the cooked flavour of heated milk from sulfur compounds released from whey proteins, the buttery note of diacetyl, the light-induced flavour from methional, and the cardboard flavour of oxidised milk from aldehydes. Many of these have odour thresholds in the parts-per-billion range.
Thresholds describe sensitivity. The absolute or detection threshold is the lowest concentration at which a stimulus is detected; the recognition threshold is the lowest at which it can be identified; the difference threshold, or just-noticeable difference, is the smallest detectable change; and the terminal threshold is the level above which further increases are not perceived. Weber's law states that the just-noticeable difference is a roughly constant fraction of the starting intensity, ΔI/I = k, and Stevens' power law relates perceived intensity to stimulus concentration as a power function. Repeated exposure leads to adaptation, a temporary loss of sensitivity, which is why judges rest and rinse between samples.
Appearance is judged first and shapes expectations. Colour, gloss, uniformity, the presence of sediment, cream line, wheying off, mould or openness in cheese are seen before tasting, and colour strongly influences perceived flavour: coloured drinks are often judged to taste of the fruit their colour suggests, regardless of the actual flavour. For this reason, panel booths often use coloured or dim lighting to mask colour differences when colour is not being judged.
Touch and pressure receptors in the mouth, lips and hands sense texture, the mechanical and geometric properties of a food: firmness, smoothness, graininess, viscosity, creaminess, and the coolness of melting fat or ice. Texture terms such as body, mouthfeel and consistency are central to dairy judging. Hearing contributes for crisp inclusions and for the sound of breaking chocolate coatings.
Judges are subject to psychological errors that well-designed tests guard against. The expectation error arises when the judge knows or guesses something about a sample; samples are therefore coded with random three-digit numbers. The stimulus error arises from irrelevant cues such as differences in container or temperature. The logical and halo errors link unrelated attributes, as when a judge who sees a yellow butter scores its flavour higher. Positional and order errors depend on where or when a sample is presented, which is why presentation order is randomised or balanced. The contrast error makes a sample seem worse after a very good one; central tendency makes inexperienced judges avoid the ends of a scale; and fatigue and lack of motivation reduce accuracy as sessions lengthen.
Sensory tests use two kinds of assessors. Analytical tests, which ask whether products differ and how, use small panels of selected, screened and trained assessors who act as measuring instruments. Affective tests, which ask whether people like a product, use untrained consumers who represent the target market. Using a trained panel to predict consumer liking, or consumers to describe subtle attributes, is a common mistake. Expert graders, individuals with long experience of a product such as a cheese grader, are a third type, important in industry and in judging contests.
Tests are carried out in a controlled environment. The international standard on the design of test rooms, ISO 8589, describes separate booths that prevent assessors from seeing each other's reactions, neutral colours, controllable lighting, freedom from odours and noise, comfortable temperature and humidity, and a separate preparation area. Samples are presented in identical, odour-free containers, at a controlled temperature appropriate to the product, in equal portions, coded with random three-digit numbers, and in randomised or balanced order. Water and unsalted crackers or plain bread are provided to cleanse the palate. The number of samples per session is limited to avoid fatigue, particularly for strongly flavoured or fatty products such as cheese and butter, where carry-over between samples is high.
Assessors for a trained panel are selected for availability, interest, good health, absence of allergies or dietary restrictions relevant to the products, and the ability to concentrate and communicate. The international standard ISO 8586 sets out procedures for the selection, training and monitoring of assessors.
Screening tests then check sensory ability. Basic taste tests ask candidates to identify solutions of sucrose, citric or lactic acid, sodium chloride, caffeine or quinine and monosodium glutamate at moderate concentrations, and to rank them by intensity. Odour recognition tests present common odorants and, for dairy panels, defect references such as diacetyl, an oxidised milk and a rancid milk. Discrimination tests with the actual products, such as triangle tests between milks differing slightly in fat or with a low level of added off-flavour, show whether candidates can detect the differences that matter. Candidates who consistently fail are not selected; about a third to a half of candidates are commonly rejected.
Screening also identifies specific anosmias, the inability to smell particular compounds, which are surprisingly common; a candidate who cannot smell a key defect compound is unsuitable for a panel assessing that defect, though otherwise capable.
Training develops a shared vocabulary and consistent use of scales. For defect judging, assessors are trained with reference samples prepared to show each defect: oxidised milk made by adding a trace of copper sulfate and storing cold; light-induced flavour by exposing milk in clear bottles to sunlight or fluorescent light for a short time; rancid flavour by agitating raw milk, which activates lipoprotein lipase, or by adding a little butyric acid; cooked flavour by heating milk; acid flavour by adding a small amount of cultured buttermilk; feed and foreign flavours by adding traces of the relevant material. For descriptive panels, assessors agree on attribute definitions, references for the ends and midpoints of scales, and the evaluation procedure, over many sessions; descriptive training often takes 40 to 120 hours or more depending on the product and method.
Panel performance is monitored continuously. For each assessor, three properties are checked: discrimination, the ability to distinguish products that differ; repeatability, giving the same score to the same sample in replicate; and agreement or consistency with the panel, scoring in the same direction as others. These are evaluated by analysis of variance of replicated data, with product, assessor, replicate and their interactions as terms. A significant product by assessor interaction may show that assessors disagree on the direction of differences, or simply use the scale differently. Assessors who drift are retrained, and the panel is recalibrated with reference samples at intervals. Motivation is maintained by feedback, variety and recognition.
Discrimination tests answer the question: is there a perceptible difference between two products? They are used when a process, ingredient, supplier or package is changed and the producer wants to know whether consumers could notice, when a new line or plant must match an existing product, and to screen assessors. They are forced-choice tests: the assessor must give an answer even if guessing, so that the number of correct answers expected by chance is known exactly and can be compared with the number observed using the binomial distribution.
The probability of a correct answer by chance, p₀, depends on the test design. If the observed number of correct responses is significantly higher than expected by chance, it is concluded that a perceptible difference exists. The number of correct answers needed for significance at a chosen level depends on the number of assessors and on p₀; standard tables, such as those in the ISO standards for each test, give these minimum numbers, or they can be calculated from the binomial distribution. For a triangle test with 30 assessors, at least 15 correct answers are needed for significance at the 5% level.
A non-significant result does not prove that the products are the same, especially with few assessors. To demonstrate similarity, the test is designed with enough assessors to control the risk β of missing a difference of a specified size, often expressed as the proportion of distinguishers, pd, the fraction of the population able to detect the difference. The proportion of distinguishers is estimated from the proportion correct, pc, by pd = (pc − p₀)/(1 − p₀).
In the triangle test (ISO 4120), each assessor receives three coded samples, two identical and one different, and must identify the odd one. With six possible orders (AAB, ABA, BAA, BBA, BAB, ABB), presented in balanced fashion, the chance probability is one in three. The test is efficient in assessor numbers but demanding, because three samples must be tasted and compared, and it suffers from fatigue and carry-over with strongly flavoured products such as cheese.
In the duo-trio test (ISO 10399), the assessor first receives a reference sample, then two coded samples, one of which matches the reference, and must say which. The chance probability is one half, so more assessors are needed than for a triangle test to detect the same difference, but the task is simpler. In the constant-reference version the reference is always the current product; in the balanced version each product serves as reference equally often.
The paired comparison test (ISO 5495) presents two samples and asks a specific question about one attribute: which is sweeter, saltier or more bitter? This directional version, also called the two-alternative forced choice (2-AFC) test, has a chance probability of one half and is a one-sided test when the expected direction is known in advance. It is very sensitive because it focuses attention on one attribute, but it is useful only when the nature of the difference is known. A paired preference test, which asks which sample is liked more, is an affective test with consumers and is usually two-sided.
Several other designs are used. The tetrad test presents four samples, two of each product, and asks the assessor to sort them into two groups of two; its chance probability is one third, like the triangle, but it is statistically more powerful than the triangle for the same number of assessors because the cognitive task is easier to perform consistently. The two-out-of-five test asks the assessor to identify two samples of one product among five and has a chance probability of one in ten, making it efficient but very fatiguing. The A-not-A test familiarises assessors with product A and then presents samples one at a time, asking whether each is A or not A; it suits products that cannot be presented together. The same–different test presents a pair and asks whether they are the same or different.
Thurstonian models explain why tests with the same chance probability differ in power. They assume that each sample produces a perceived intensity that varies from tasting to tasting, and they express the size of a sensory difference as d′ (d-prime), the distance between the two product distributions in standard deviation units. The same d′ produces different proportions of correct answers in different tests, depending on the decision rule the assessor must use: for a given d′, the 2-AFC and tetrad tests give higher proportions correct than the triangle or duo-trio. Expressing results as d′ allows comparison across tests and studies.
In practice, dairy companies choose tests by purpose: triangle or tetrad tests for overall difference after process changes, directional paired comparisons for a known attribute such as sweetness after sugar reduction, and similarity designs when the aim is to show that a cheaper ingredient makes no noticeable difference.
Descriptive analysis measures the kind and intensity of a product's sensory attributes with a trained panel, producing a sensory profile. Several methods exist. The flavour profile method, one of the earliest, uses a small expert panel reaching consensus on flavour notes and their intensities. Quantitative descriptive analysis, introduced by Stone and Sidel, uses about 8 to 12 trained assessors who develop their own vocabulary, rate each attribute independently on unstructured line scales, commonly 15 cm long with anchors near the ends, and whose replicated scores are analysed statistically. The Spectrum method uses standardised lexicons and absolute intensity scales anchored by reference foods, so that results can be compared across products and panels. Texture profile analysis by panels rates mechanical, geometrical and fat- and moisture-related texture attributes in the order in which they are perceived.
Lexicons, agreed lists of attributes with definitions and references, have been developed for many dairy products, including cheddar-type and other cheeses, milk, whey proteins, yoghurt and butter. A cheese lexicon may include basic tastes, cooked, whey, diacetyl, milkfat, sulfur, brothy, nutty, catty, fruity and others, each defined with a reference. Such lexicons let research groups and companies in different countries describe flavours in the same terms.
Rapid descriptive methods have spread because full training is slow and expensive. Check-all-that-apply questions ask consumers or semi-trained assessors to tick attributes from a list; projective mapping (napping) asks assessors to place samples on a sheet so that similar products are close together; flash profiling lets each assessor use their own terms; and temporal methods, such as time–intensity and temporal dominance of sensations, record how perception changes during eating, important for products like cheese and ice cream whose flavour develops as they melt.
Affective tests measure liking or preference, and must use consumers who represent the target market. The most widely used scale is the nine-point hedonic scale developed by Peryam and Pilgrim in the 1950s, running from 'dislike extremely' (1) through 'neither like nor dislike' (5) to 'like extremely' (9). It is easy for consumers to use and gives data usually analysed by analysis of variance. Variants include shorter category scales, labelled magnitude scales and face scales for children.
Preference tests ask consumers to choose. A paired preference test asks which of two samples is preferred, and is analysed as a two-sided binomial test with p = 1/2, often with a 'no preference' option handled by agreed rules. Ranking tests ask consumers to order several samples by preference and are analysed with Friedman's test. Preference tells which product wins but not by how much or whether either is liked; hedonic ratings give that information.
Diagnostic questions help explain liking. Just-about-right scales ask whether an attribute such as sweetness, saltiness or thickness is too little, just about right or too much; penalty analysis then estimates how much overall liking falls among consumers who find an attribute not just right, showing which changes would improve the product most. Purchase-intent questions and willingness-to-pay methods connect sensory liking to market behaviour, though stated intentions overestimate real purchases.
Trained panels should not be asked about liking, and consumers should not be asked to rate subtle technical attributes they cannot reliably perceive; each kind of panel answers the kind of question it is suited to.
The first step in any sensory study is to define the question precisely, because the question determines the method. Is there any difference? A discrimination test. Is product A sweeter than B? A directional paired comparison. What is the flavour and texture of this cheese and how does it change during ripening? Descriptive analysis. Which recipe do target consumers like best? A hedonic test with consumers. Does this lot meet the grade standard? A scorecard evaluation by qualified graders.
The second step is to choose assessors suited to the question, determine how many are needed for the required power, design the presentation (number of samples, order, coding, temperature, portion), and plan the analysis before collecting data.
Common errors in dairy companies include using a few managers tasting openly in an office as a sensory panel; asking a trained panel whether a product is liked; drawing conclusions about similarity from a non-significant triangle test with ten assessors; serving cold and warm samples together; and presenting samples in the same order to everyone, so that order effects are confused with product differences. Good sensory practice is less about complex statistics than about these basic disciplines, which turn opinion into evidence.
Scorecards are the traditional tools of dairy grading and judging. A scorecard divides a product's quality into a few categories, typically flavour, body and texture, colour and appearance, and for some products salt, finish, melting quality or package, assigns each a maximum score, and lists the defects that reduce the score in each category, often with guidance on how many points each defect at a given intensity should cost. The total score, or a grade based on it, summarises quality. Because flavour is the most important attribute to consumers and the most sensitive indicator of problems, it usually carries the greatest weight.
Scorecards serve several purposes: grading products for payment or export, routine quality control in plants, competitions and shows, and the training and examination of students. Their weaknesses are that the weights are fixed by convention rather than by measured consumer importance, that experts may disagree on the severity of defects, and that a high score means freedom from defects rather than consumer liking: a product with no defects can still be bland and unappealing. Modern quality programmes therefore often combine scorecard grading with descriptive and consumer testing.
International standardisation exists. The ISO and IDF standard for sensory analysis of milk and milk products (ISO 22935, IDF 99) sets out general guidance on recruiting and training assessors, recommended methodology, and a method of evaluating compliance with product specifications using a five-point scale for each attribute, such as appearance, consistency, odour and flavour, from no deviation from the specification down to very large deviation, with product-specific weighting factors. National grading schemes and student judging programmes in many countries use their own scorecards with different point allocations.
In milk judging, flavour dominates. Judges sniff the sample, take a small sip, roll it around the mouth to warm it and bring volatiles to the nose retronasally, and expectorate, noting first impressions and aftertaste. A typical milk scorecard grades flavour on a scale on which a sample with no criticism receives the top score and each defect deducts points according to its intensity, together with a smaller allowance for appearance such as sediment and cream line. The defects scored include cooked, feed, flat, foreign, lacks freshness, light-induced, malty, oxidised, rancid, salty, acid, bitter, fruity and unclean, described in the next chapter. Samples are judged at a moderate temperature, slightly warmer than refrigeration, because volatile defects are hard to detect in very cold milk.
Butter scorecards give the largest share to flavour, with smaller shares for body and texture, colour, salt and sometimes package and finish. Body and texture are assessed by cutting and plugging the butter with a trier, observing the cut surface for free moisture, bending and breaking the plug, and judging spreadability and mouthfeel; defects include crumbly, greasy, gummy, leaky, mealy, short and sticky. Colour should be uniform; mottled or streaked butter indicates uneven salt or moisture distribution. Salt is judged for amount and for grittiness from undissolved crystals.
Ghee, the clarified butter fat of South Asia and the Middle East, is judged on flavour, which should be the characteristic cooked, nutty aroma developed during clarification; on texture, especially the formation of uniform granules on slow cooling; on colour, which depends on the species of milk and the clarification temperature; and on freedom from sediment and moisture. Common defects are burnt or scorched, rancid, oxidised or tallowy, cheesy or sour, and smoky flavours, and a greasy, non-granular texture.
Cheese judging, for varieties such as cheddar-type cheeses, assesses flavour, body and texture, and colour and appearance, and for some scorecards finish. The cheese is sampled with a trier, and the plug is examined for colour uniformity, openness and mechanical holes, then smelled immediately, worked between fingers for body and texture, and tasted. Flavour defects include acid, bitter, fruity, flat, unclean, sulfide, rancid, yeasty, whey taint and feed flavours; body and texture defects include crumbly, curdy, corky, mealy, pasty, short, weak, gassy and open. Ripened cheeses are judged against the expected character of their age and variety, so a mild young cheese is not penalised for lacking the intensity of a mature one. Cheese certifications for retail professionals test the ability to evaluate cheese against its variety's standard and to recognise defects and their causes.
Ice cream is judged on flavour, body and texture, melting quality, colour and package. Flavour is assessed for the intensity and naturalness of the characterising flavour, sweetness and the absence of defects such as cooked, oxidised, lacks fine flavour, old ingredient, rancid, salty and storage flavours. Body and texture are assessed as the ice cream melts in the mouth: coarse or icy, sandy, crumbly, fluffy, gummy, soggy and weak are defects. A sample is also allowed to melt on a plate and its meltdown observed for foamy, curdy or wheyed-off melting.
Yoghurt and other fermented milks are judged on flavour (balance of acidity and aroma, absence of yeasty, bitter, unclean or chemical notes), body and texture (smoothness and firmness appropriate to set or stirred styles; defects such as wheying off, weak, lumpy, grainy or ropy) and appearance. Fruit yoghurts are also judged on the quality and distribution of fruit.
Many flavour defects arise before milk leaves the farm. Feed flavours are absorbed from strong-smelling feeds, such as silage, certain pastures, brassicas, onions, garlic and some weeds, when cows eat them shortly before milking; volatile compounds pass from the digestive tract or lungs into the blood and then into the milk within an hour or two. They are reduced by feeding such feeds after milking rather than before. Barny and cowy flavours come from poorly ventilated barns, where milk and cows absorb odours, and from ketosis in cows, which raises acetone and other ketone bodies in milk.
Salty milk occurs with mastitis and late lactation, when the udder's barriers break down and sodium and chloride rise while lactose falls. Unclean, foreign or chemical flavours arise from contamination with cleaning and sanitising chemicals, medicines, fuels or disinfectants, and from absorbed odours of paint or fuel near the milk. Phenolic and medicinal taints can be caused by chlorophenols formed when chlorine sanitisers react with phenolic compounds in water or on equipment, detectable at extremely low concentrations.
Rancid flavour, described as soapy, bitter or like butyric acid, results from lipolysis, the release of free fatty acids, especially the short-chain butyric and caproic acids, by lipoprotein lipase. The native enzyme is mostly bound to casein micelles and separated from the fat by the milk fat globule membrane, but damage to the membrane by excessive agitation, foaming, pumping of warm milk, freezing and thawing, or homogenisation of raw milk exposes the fat to the enzyme; some cows, in late lactation or on poor diets, give spontaneously lipolytic milk. Pasteurisation inactivates lipoprotein lipase, so rancidity in pasteurised milk usually began before heating or comes from heat-stable bacterial lipases.
Oxidised flavour, described as cardboard, papery, metallic or tallowy, results from oxidation of unsaturated fatty acids, especially in the phospholipids of the fat globule membrane, to hydroperoxides that break down into aldehydes such as hexanal, heptanal and unsaturated aldehydes with very low thresholds. Copper and iron catalyse it, so contamination from copper-containing equipment or water was once a common cause; it is favoured by low temperature storage, high dissolved oxygen and diets rich in polyunsaturated fat or low in vitamin E, and is reduced by antioxidants in the milk, such as tocopherols, and by avoiding metal contamination.
Light-induced flavour, also called sunlight or activated flavour and described as burnt, burnt-feather or cabbage-like, develops within minutes to hours when milk is exposed to sunlight or fluorescent light in clear or translucent packaging. Riboflavin absorbs light and acts as a photosensitiser, producing reactive oxygen species and oxidising the sulfur amino acid methionine to methional, the key compound, as well as initiating lipid oxidation. It is prevented by opaque or light-blocking packaging and by protecting milk from light in retail display. It differs from metal-induced oxidised flavour in its speed and its burnt, sulfurous character, though prolonged light exposure also produces cardboard notes.
Heat produces its own flavours. Cooked flavour, slight in pasteurised milk and stronger in milk heated more severely, comes from sulfur compounds, such as hydrogen sulfide and methanethiol, released from the thiol groups of denatured β-lactoglobulin and from fat globule membrane proteins; the sulfurous, cabbagey notes of fresh UHT milk fade over days as these compounds are oxidised. Caramelised and stale flavours of heated and stored UHT milk come from Maillard reaction products and methyl ketones, such as 2-heptanone and 2-nonanone, formed from fat on heating. Scorched or burnt flavour indicates fouling or overheating.
Bacteria produce many flavour defects. Acid or sour flavour develops when lactic acid bacteria grow in milk that has been poorly cooled. Malty flavour, resembling malted milk or grapenuts, is produced by certain strains of Lactococcus lactis, historically called variety maltigenes, which form 3-methylbutanal from leucine; it also indicates poor cooling. Fruity or fermented flavours, like pineapple or strawberry, are produced by psychrotrophic bacteria such as Pseudomonas fragi, which form ethyl esters of short-chain fatty acids during prolonged cold storage. Bitter flavour develops when proteases from psychrotrophs, or from spore-formers such as Bacillus cereus in pasteurised milk, hydrolyse casein into bitter peptides. Unclean, putrid or stale flavours accompany high bacterial numbers generally. Most bacterial defects become detectable only when counts reach millions per millilitre, so their appearance signals advanced spoilage.
Several defects relate to composition or processing. Flat or watery flavour indicates low solids, as in milk diluted with water or of naturally low composition. Lacks freshness or stale describes milk that is not defective in any specific way but has lost the clean, sweet character of fresh milk, an early sign of age. Chalky or astringent mouthfeel may follow high heat treatment or reconstitution. Absorbed flavours, such as refrigerator odours from onions, fish or fruit, enter through permeable packaging or open containers.
A trained judge identifies a defect, estimates its intensity and, crucially, connects it to its likely cause, so that the farm, the transport or the plant can act. Defect identification thus links the sensory laboratory directly to quality control.
The body and texture of butter depend on the proportion of solid to liquid fat at the serving temperature, the size and arrangement of fat crystals, the dispersion of water droplets and the working of the butter. Greasy butter is too soft and oily, from a high proportion of liquid fat, as in summer pasture milk fat, or from overworking; sticky butter clings to the trier and the palate, often from fat high in high-melting triglycerides or from cream treatment; gummy butter feels waxy and chewy, linked to fats rich in high-melting glycerides, as in some winter feeds; crumbly butter breaks into pieces when cut, from a high solid fat content, insufficient working or very hard winter fat; mealy or grainy butter has a granular mouthfeel from coarse fat crystals, often caused by poor tempering of cream or by neutralised high-acid cream. Leaky butter shows beads of free water on the cut surface, caused by insufficient working, which leaves the moisture poorly dispersed; it promotes microbial growth. Mottled or streaked colour indicates uneven salt distribution. Physical ripening programmes of cream, which control crystallisation by temperature cycling, are the main means of adjusting butter consistency to the season.
Ghee should form fine, uniform granules on slow cooling; a greasy, non-granular texture results from rapid cooling or incorrect composition. Cream defects include feathering in coffee, a lumpy body from partial churning during handling, and a thin body from low fat or poor ageing.
Cheese texture is set by its composition, especially moisture, fat, salt and pH, by the state of the casein network and its calcium, and by proteolysis during ripening. Common defects are described in standard terms. Crumbly cheese breaks into small pieces; it is associated with low pH, over-acid cheese, low moisture or excessive salt. Curdy or rubbery cheese is springy and resists breakdown, typical of young cheese with intact protein. Corky cheese is dry, hard and resists pressure, from low moisture or fat. Pasty or sticky cheese is soft and smears, from high moisture or high acid with breakdown of the protein network. Mealy cheese has a grainy, floury mouthfeel, sometimes from high acid or high heat. Short cheese breaks without bending, from high acid. Weak or soft cheese lacks resistance, from high moisture or excessive proteolysis. Open texture with irregular mechanical holes results from poor knitting of curd particles; gas holes in cheeses that should be closed indicate early blowing by coliforms or late blowing by clostridia; slits and cracks may also arise from gas.
For varieties with expected eyes, such as Swiss-type cheeses, the size, number, distribution and gloss of the eyes are judged, and blind cheese without eyes is a defect. Processed cheese is judged for smoothness, sliceability, melt and the absence of graininess, which can come from crystals of emulsifying salts or calcium phosphate.
Ice cream texture is the perception of the size, shape and distribution of ice crystals, air cells, fat globule aggregates and lactose crystals. Coarse or icy texture, the commonest defect, comes from large ice crystals, which grow during storage when temperatures fluctuate (heat shock) or when the mix has too little solids or stabiliser. Sandy texture, a gritty feel that does not melt away, comes from lactose crystals, likely when milk solids-not-fat are high and storage is long. Crumbly texture indicates low solids, low fat or poor stabilisation; fluffy or spongy ice cream has too much air, an excessive overrun; gummy or doughy ice cream has too much stabiliser; and soggy or heavy ice cream has too little air. Melt-down defects include a foamy melt from excessive overrun or emulsifier, a curdy melt from protein instability, and wheying off.
Yoghurt and fermented milks show wheying off or syneresis, a layer of free whey caused by low solids, insufficient heat treatment of the milk, excessive acidity, disturbance during setting, or high incubation temperature; lumpy or grainy texture from poor mixing or rapid acidification; weak, thin body from low solids or poor culture activity; and ropy or slimy texture, desirable in some Nordic products but a defect elsewhere, from exopolysaccharide-producing strains.
Concentrated and dried products have their own defects. Sweetened condensed milk can become sandy from large lactose crystals when seeding and cooling are not controlled, or can thicken during storage (age thickening). Milk powders can be lumpy or caked from moisture uptake, can show poor wettability or solubility, and can contain scorched particles from overheating in the dryer.
Discrimination data are counts of correct and incorrect answers, analysed with the binomial distribution. For n assessors and chance probability p₀, the probability of x or more correct answers by chance is the upper tail of the binomial distribution. The test is one-sided for triangle, duo-trio, tetrad and directional paired comparison tests, because only more correct answers than chance indicate a difference. For a triangle test, with p₀ = 1/3, the minimum numbers correct for significance at the 5% level are 8 of 12, 10 of 18, 13 of 24, 15 of 30 and 18 of 36; for duo-trio and directional paired comparison tests, with p₀ = 1/2, they are 10 of 12, 13 of 18, 17 of 24 and 20 of 30. With large numbers, a normal approximation with z = (x − np₀)/√(np₀(1 − p₀)) is used, often with a continuity correction.
Paired preference tests are two-sided, because either product might be preferred, and need slightly more responses for significance than one-sided tests at the same level. 'No preference' responses are either excluded, split equally between the two products, or analysed separately, and the rule must be decided in advance.
Replicated discrimination tests, in which each assessor does several trials, violate the assumption of independent observations if assessors differ in ability; beta-binomial models account for this. Power and sample size calculations for discrimination tests use the binomial distribution under an alternative hypothesis, such as a stated proportion of distinguishers, and are essential when the goal is to show similarity.
Scores on descriptive and hedonic scales are analysed with analysis of variance. In a descriptive study, each attribute is analysed with a model including products, assessors, replicates and the product by assessor interaction. Assessors are usually treated as a random effect, representing a population of possible assessors, and the product effect is then tested against the product by assessor interaction rather than the residual error, a point often overlooked. Significant product differences are followed by multiple comparison procedures, such as Tukey's honestly significant difference or Fisher's least significant difference, to find which products differ.
Hedonic data from consumers are analysed similarly, with consumers as a random blocking factor. Consumer populations are often not homogeneous: groups of consumers may like different products, and averaging hides this. Cluster analysis of individual liking scores identifies segments with different preferences, which is valuable in product development.
Ranked data, from ranking tests, are analysed with the Friedman test, a non-parametric analysis for blocks of ranks, followed by multiple comparisons of rank sums. Data from just-about-right scales are summarised as percentages in each category and analysed with penalty analysis, and check-all-that-apply data with Cochran's Q test and correspondence analysis.
Assumptions matter. Sensory scales are bounded and assessors use them differently, so distributions may be skewed and variances unequal; transformation, non-parametric methods or models that account for scale use are used when needed. Above all, the experimental design, randomisation and replication must be sound; no analysis can rescue a poorly designed test.
Sensory profiles contain many correlated attributes, so multivariate methods are used to summarise them. Principal component analysis of the product mean scores reduces many attributes to a few components and displays products and attributes on a map: products close together have similar profiles, and attributes pointing in the same direction are correlated. A PCA of cheese profiles during ripening, for example, typically shows products moving along a component associated with increasing intensity of brothy, sulfur and nutty flavours and decreasing milky and whey notes.
Generalised Procrustes analysis aligns the configurations of individual assessors who used different scales or vocabularies, as in free-choice and flash profiling, before finding the consensus. Multiple factor analysis combines several blocks of data, such as sensory, instrumental and consumer data, in one analysis.
Preference mapping relates consumer liking to product characteristics. Internal preference mapping performs PCA on the consumers' liking scores, showing the directions of preference of individual consumers in the product space; external preference mapping regresses each consumer's liking on the sensory dimensions from descriptive analysis, showing which sensory characteristics drive liking. Together they tell developers which sensory direction to move a product to please a target segment. As with all multivariate methods, results depend on the products included, and maps should be interpreted with knowledge of the products rather than mechanically.
Instruments complement sensory panels by giving fast, objective and reproducible measurements, but they measure physical or chemical properties, not perception, so their relation to sensory attributes must be established, not assumed. Texture analysers compress, penetrate, cut or extend samples and record force against deformation. Texture profile analysis by instrument compresses a sample twice in succession, imitating two bites, and derives hardness (peak force in the first compression), cohesiveness (ratio of the work of the second compression to the first), springiness (recovery of height between compressions), adhesiveness (negative work to pull the probe away) and derived parameters such as chewiness. Penetration tests measure the firmness of butter and set yoghurt; rheometers measure viscosity and viscoelastic properties of yoghurts, creams and cheese, including melt behaviour on heating.
Correlations between instrumental and sensory texture are often good for attributes that are mainly mechanical, such as firmness, and poorer for attributes that integrate many factors, such as creaminess, which depends on viscosity, fat, particle size and lubrication. Colour is measured with colorimeters in the CIE L*a*b* system: L* for lightness, a* for red–green and b* for yellow–blue; butter and cheese colour, browning of heated milk products and whiteness of powders are routinely measured this way. Instrumental colour usually correlates well with visual assessment under standard lighting.
Gas chromatography separates the volatile compounds of a food and, coupled with mass spectrometry, identifies and quantifies them. Hundreds of volatiles may be found in a cheese, but only a fraction contribute to its aroma. To find those that matter, gas chromatography–olfactometry splits the column effluent between a detector and a sniffing port, where a trained assessor records the odours perceived as compounds elute. Aroma extract dilution analysis repeats the sniffing on serial dilutions of the extract; compounds still detectable at high dilution are the most potent odorants.
The odour activity value relates concentration to perception: OAV = concentration in the food divided by the odour threshold in a relevant medium. Compounds with values above one are likely to contribute to aroma. For example, a compound present at 20 µg/kg with a threshold of 2 µg/kg has an OAV of 10. Reconstitution of a food's aroma from its key compounds, and omission experiments in which one compound is left out, confirm their importance.
In milk and dairy products, such work has identified key compounds for many flavours: hexanal and other aldehydes as markers of oxidation; methional for light-induced flavour; dimethyl sulfide and other sulfur compounds for cooked notes; methyl ketones such as 2-heptanone for the stale flavour of UHT milk; diacetyl for the buttery note of cultured products; acetaldehyde for yoghurt; and many compounds from amino acid and fat breakdown in cheese. Electronic noses and tongues, arrays of sensors with pattern recognition, can classify samples rapidly, for example detecting off-flavoured milk, but they do not measure perception and must be calibrated against sensory data for each application.
Statistical models link instrumental measurements to sensory scores so that instruments can predict sensory properties in routine control. Simple linear regression suffices when one instrumental measure tracks one attribute closely, such as penetrometer firmness and sensory firmness of butter. More often many instrumental variables relate to many sensory attributes, and partial least squares regression is used, building latent variables from the instrumental data that best predict the sensory data. Models are validated on samples not used to build them, and their prediction errors are compared with the variation that matters.
Such models are valuable but limited. They are valid only for the range of products, processes and storage conditions used to develop them; a model built on young cheddar-type cheeses may fail for mature ones. Instrumental measures may miss perceptual interactions, such as the way sweetness masks bitterness or fat reduces the perception of some volatiles. And a model predicting a trained panel's ratings does not predict consumer liking unless consumer data are also included.
In practice, dairy plants use instruments for routine screening, such as colour of butter, firmness of cheese, viscosity of yoghurt and volatile markers of oxidation, and keep sensory evaluation for release decisions, defect investigation and product development, where the integrated judgement of trained people remains indispensable.
Consumer tests measure the reactions of the people who buy and eat a product. They are conducted in central locations, such as shopping centres or test facilities, where conditions are controlled; in homes, where products are used naturally over days but conditions vary; or online with products delivered. Each setting trades control against realism.
Consumers are recruited to represent the target market, usually by age, sex, region and, most importantly, by use of the product category: a test of a new cheese recruits people who buy cheese. Numbers must be large enough for reliable conclusions; tests commonly use 75 to 150 or more consumers per segment, more when differences are small or segments are to be compared. Each consumer evaluates samples presented one at a time in balanced order, with rest and palate cleansing between them. Questionnaires usually start with overall liking, before more detailed questions can bias it, and then ask about the liking of appearance, flavour and texture, just-about-right attributes, purchase intent and open comments.
Consumer tests are used to choose between prototypes, to check that cost-saving changes do not reduce liking, to compare a product with competitors, and to set sensory specifications. They should not be replaced by internal tasting by company staff, who are familiar with the product and not representative. Ethics require informed consent, attention to allergies and dietary needs, and food safety of test products.
The shelf life of many dairy products is limited not by safety but by sensory change: oxidised flavour in milk powders and UHT milk, bitterness and age gelation in UHT milk, souring of pasteurised milk, excessive acidity in yoghurt, staling of cream, flavour loss or rancidity in butter, and texture changes in ice cream and cheese. Sensory shelf life is the time during which the product remains acceptable, and determining it requires a criterion, the sensory end point.
Several approaches define the end point. A trained panel may follow the intensity of a key defect over time, and the end point is set at an intensity known to be unacceptable; or a difference-from-control test may be used, the end point being reached when the stored product differs significantly or by a set amount from a fresh control. These methods are sensitive but do not by themselves show when consumers would reject the product.
Consumer-based methods ask whether the product is acceptable. In survival analysis, developed for food shelf life by Hough and colleagues, consumers taste samples stored for different times and say whether they would normally consume each; the proportion rejecting is modelled as a function of storage time, and the shelf life is set at the time when a chosen proportion of consumers, such as 25% or 50%, would reject the product. The choice of proportion is a business decision. In practice, companies combine methods: trained panels or instruments track change routinely, and consumer data calibrate the end point.
A shelf-life study requires samples representative of normal production, stored under the conditions they will meet in distribution and in the home, including realistic temperature abuse where relevant, and sampled at enough time points to define the change. A difficulty is that samples stored for different times would normally be tasted on different days, so day-to-day variation of the panel is confused with storage effects. The reversed storage design avoids this: samples are placed into storage at different dates, or moved from a stable condition such as frozen storage into the test condition at staggered times, so that all ages are ready on the same day and are evaluated together in one session.
Accelerated shelf-life tests store products at higher temperatures to speed deterioration, and extrapolate to normal conditions using kinetic models such as the Arrhenius equation or Q10 factors. They are useful for slow-changing products such as milk powders and UHT milk, but the mechanism of deterioration can change at higher temperatures, so accelerated results must be confirmed by real-time studies.
Sensory, microbiological and chemical data should be considered together. A pasteurised milk may become unacceptable through bacterial off-flavours before counts exceed specifications, or reach count limits before flavour changes; a milk powder's oxidation may be detected by a trained panel before hexanal measurement shows a clear rise. Shelf-life claims on packs, use-by or best-before, must take account of food safety first; sensory end points apply to quality, within the limit set by safety.
Train your senses deliberately. Prepare the classic defect references, oxidised milk with a trace of copper, light-induced milk in a clear bottle on a windowsill, rancid milk by shaking warm raw milk, cooked milk by boiling, and taste them blind until you can name each at low intensity. Take part in judging practice for milk, butter, cheese, ice cream and yoghurt with a scorecard. Learn the chance probabilities and the binomial minimum numbers for discrimination tests, and run a real triangle test with your class to see how often guessing succeeds.
A quality or product development technologist five years into the job runs the plant's sensory programme: daily grading of milk and products against specifications, triangle or tetrad tests after process and supplier changes, investigation of consumer complaints by identifying the defect and tracing its cause, panel screening and monitoring, and shelf-life studies with defined end points. Typical decisions: whether a new packaging film causes a detectable flavour change, whether a lot with a slight oxidised note can be released, and how many assessors are needed to show that a cheaper stabiliser makes no difference.
A head of quality or of research and development with 25 years' experience decides how sensory evidence is used in the business: which products are released on sensory grounds, how consumer testing guides reformulation for sugar, salt and fat reduction, how shelf-life claims are set, and how sensory specifications are written into supplier and customer contracts. They protect the independence of the sensory function from commercial pressure, because a panel that is overruled whenever its results are inconvenient soon stops being useful.
A professor or research leader works on the chemistry of dairy flavour and its perception, sensory lexicons and methods, the drivers of consumer liking across cultures and markets, temporal perception of melting products, and models linking instrumental and sensory data. Research quality depends on sound design and statistics, on reporting panel training and performance, and on representative consumer samples. Teaching judging and sensory science together keeps the craft knowledge of experienced graders alive while placing it on a scientific footing.
Exact point allocations of national and collegiate scorecards are described in outline only, because they differ between organisations and editions; graders should use the current card of their scheme. Thresholds of individual aroma compounds are not quoted because they depend strongly on the medium and method. Detailed Thurstonian calculations, beta-binomial models, cross-cultural consumer research and the neuroscience of flavour are introduced but not worked through. Product-specific defect causes are covered in more depth in the product technology subjects.
Original practice questions written to the syllabus and pattern of the exams named in each question's examTags. They are not past papers and do not reproduce any board's questions.
These notes are free to read here, with no account.
What the Laisha app adds:
Free for ten days, then a subscription. We would rather write that here than have you find it later.