Food Science & Technology · Food Science Core · also Sensory Evaluation of Foods, Organoleptic Evaluation, Judging of Food Products, Consumer Sciences
Measuring how people perceive food: the physiology of the senses, panel selection and training, difference, descriptive and affective tests, the statistics that analyse them, and consumer research for product development.
In the exams: ICAR AIEEA-PG Code 18 Unit III (judging of food products); ICAR AICE 35.a Unit 4 (sensory evaluation, panel screening)
Sensory science measures how people perceive food through sight, smell, taste, touch and hearing, and turns those perceptions into data that can guide decisions. It uses people as measuring instruments, which makes it both indispensable, because no instrument can tell whether a product tastes good, and difficult, because people vary, tire, adapt, are influenced by expectations and can be biased by the way a test is run. The discipline therefore combines physiology and psychology with careful test design and statistics.
Sensory and consumer tests answer three kinds of question. Discrimination tests ask whether products differ, for example after a change of supplier or a salt reduction. Descriptive tests ask how products differ, using trained panels to profile attributes such as sweetness, crispness or fruity aroma. Affective tests ask whether people like a product and which they prefer, using consumers from the target market. Choosing the right kind of test for the question is the first rule of the subject; using trained panelists to predict liking, or untrained consumers to describe subtle attributes, gives misleading results.
The subject is examined in ICAR AIEEA-PG Code 18 Unit III (judging of food products) and the ICAR AICE-JRF/SRF 35.a syllabus (sensory evaluation, panel screening), and it underpins quality control, product development and research in every food company. Examiners ask for the physiology of the senses, test conditions, panel selection, the chance probabilities and uses of difference tests, scale types, descriptive methods and their originators, the hedonic scale, and the statistical treatment of sensory data. Product development applications are taught in food-product-development, and the statistics of designed experiments and multivariate methods in statistics-experimental-design-food.
Taste (gustation) is detected by taste receptor cells grouped in taste buds, located mainly in papillae on the tongue: fungiform papillae at the front, foliate papillae at the sides and circumvallate papillae at the back; filiform papillae, the most numerous, carry no taste buds and serve texture perception. Five basic tastes are widely accepted: sweet, sour, salty, bitter and umami. Sweet and umami are detected by G-protein-coupled receptors of the T1R family (T1R2 with T1R3 for sweet, T1R1 with T1R3 for umami), bitter by a large family of T2R receptors, salty mainly through sodium channels, and sour through proton-sensitive channels. The old 'tongue map', in which each taste was confined to one region, is wrong: all tastes can be detected wherever taste buds occur, though sensitivity varies somewhat. Individuals differ in taste sensitivity, for example in sensitivity to the bitter compounds phenylthiocarbamide and propylthiouracil, which divides people into non-tasters, tasters and supertasters.
Smell (olfaction) is detected by olfactory receptor neurons in the olfactory epithelium high in the nasal cavity; humans have about 400 functional olfactory receptor types, which in combination allow discrimination of a very large number of odours. Odours reach the epithelium by two routes: orthonasally, by sniffing through the nostrils, and retronasally, from the mouth through the back of the throat during eating and swallowing. Retronasal smell provides most of what people call the 'taste' of food, which is why flavour seems to vanish when the nose is blocked. Flavour is the combined perception of taste, retronasal aroma and chemesthesis, the chemical sensitivity of trigeminal nerve endings in the mouth and nose, which gives the burn of chilli capsaicin (via the TRPV1 receptor), the cooling of menthol (TRPM8), the pungency of mustard and the tingle of carbonation.
Touch in the mouth and hands provides texture: the hardness, cohesiveness, viscosity, crispness, creaminess and particle size of foods, perceived through mechanoreceptors in the lips, tongue, teeth, periodontal ligament and jaw muscles during chewing and swallowing. Temperature perception also shapes eating experience. Sight provides colour, size, shape, gloss and surface texture and creates expectations that change perceived flavour: in classic experiments, colouring a drink differently changes the flavours people report, and darker colours in fruit drinks increase perceived sweetness or flavour intensity. Hearing contributes to perceived crispness and crunchiness, through the sounds of fracture transmitted by air and bone, and altering these sounds changes judgements of freshness.
Psychophysics relates the physical intensity of a stimulus to perceived sensation. The detection or absolute threshold is the lowest concentration at which a stimulus is detected; the recognition threshold the lowest at which it can be identified; the difference threshold or just-noticeable difference the smallest change that can be detected; and the terminal threshold the level above which further increases are not perceived. Weber's law states that the just-noticeable difference is a constant fraction of the starting intensity, ΔI/I = k: if 0.5 g/L is the smallest detectable increase from 10 g/L of sugar, the Weber fraction is 0.05. Fechner proposed that sensation grows with the logarithm of intensity, while Stevens' power law, S = k × I^n, describes perceived intensity as a power function of stimulus intensity, with exponents that differ by sense and stimulus. Adaptation, the decline in sensitivity during continued exposure, and the fatigue it causes affect test design, as do interactions between tastes, such as the suppression of bitterness by sweetness.
A sensory laboratory is designed to reduce distractions and bias so that differences in responses come from the samples, not the surroundings. International guidance is given in ISO 8589 on the general design of test rooms. A typical facility has a preparation area, separated from the testing area so that cooking odours and noise do not reach assessors; individual booths for testing, so that assessors work independently without seeing or hearing each other, with a pass-through hatch for samples; a discussion or training room for panel training and consensus work; and storage for samples and materials. Booths are evenly lit, often with adjustable lighting, including coloured light such as red or green that masks colour differences when only flavour or texture is to be judged. Rooms are kept at a comfortable, controlled temperature and humidity, free of odours, often with slight positive air pressure in the testing area, and quiet. Walls and surfaces are neutral in colour. Increasingly, data are collected on computers or tablets in the booths, which randomises presentation and speeds analysis.
Samples are prepared consistently, in identical containers, at controlled serving temperatures and sizes, and presented in a way that does not reveal their identity. Each sample is labelled with a random three-digit code, because single letters or numbers such as A, B or 1 carry expectations. The order of presentation is balanced or randomised across assessors, because the first sample in a set is often rated differently (first-order or position effects) and samples influence the perception of those that follow (contrast and carry-over effects); designs such as Williams Latin squares balance both order and carry-over.
The number of samples per session is limited to avoid fatigue and adaptation, especially for strongly flavoured, spicy, fatty or alcoholic products. Palate cleansers between samples, such as water, unsalted crackers, plain bread or, for fatty foods, warm water or apple slices, and rest intervals reduce carry-over. Assessors are asked to avoid smoking, eating, strong flavours and fragrances before tests, and to report illness. Instructions and questionnaires are clear and tested in advance. These details matter because small biases, such as a warmer serving of one sample or a code that suggests a ranking, can produce statistically significant but false differences. Psychological errors that test design must counter include the error of expectation, when information about the samples leads assessors to find what they expect; the error of habituation, when assessors give the same response to a series of gradually changing samples; the stimulus error, when irrelevant cues such as sample size or container influence judgement; and the logical error, when one characteristic is assumed to go with another. Giving assessors only the information they need, presenting samples identically and randomising order are the main defences.
Sensory tests use different kinds of assessors depending on the question. Trained or expert panels, typically of about 8 to 12 selected and trained assessors, are used for descriptive analysis and some difference testing, where they act as calibrated instruments; their personal liking is irrelevant. Panels for difference tests may be screened but less extensively trained, and commonly involve about 20 to 40 or more assessors depending on the test and the risk levels chosen. Consumer panels for affective tests are untrained, drawn from the target population, and much larger, commonly at least about 60 to 100 and often more, because liking varies widely between people and the results must represent the market.
Selection of trained assessors follows ISO 8586. Candidates are recruited for interest, availability, health (no conditions or medications that affect taste and smell, no food allergies relevant to the products) and ability to communicate. Screening tests check that they can recognise basic tastes and common odours, discriminate between samples that differ slightly, for example in triangle tests, rank samples in order of intensity, and describe what they perceive. Only a fraction of candidates are usually selected.
Training develops assessors' ability to detect, recognise, describe and rate attributes consistently. Panels learn a common vocabulary of attributes, with definitions and procedures for evaluating each, and use reference standards, physical examples that anchor attributes and intensity levels, such as solutions of sucrose at defined concentrations for sweetness or specific foods for hardness. Training continues until the panel agrees on the meaning of attributes and rates samples reproducibly; for complex descriptive methods it can take many hours over weeks or months.
Panel performance must be monitored continuously. Each assessor's discrimination (ability to separate products that differ), repeatability (giving similar scores to the same sample in replicate evaluations) and agreement with the panel (consensus) are checked, usually by analysis of variance of replicated data and graphical tools. Assessors who drift, disagree or fail to discriminate are retrained or replaced. Motivation matters: assessors who are bored, rushed or unrewarded perform poorly, so feedback, varied tasks and recognition are part of panel management. Panels may be internal, drawn from company staff, which is convenient and cheap but risks bias from product knowledge and hierarchy, or external, recruited from the community and paid, which costs more but gives greater availability and independence. Staff who work on a product should not serve on panels evaluating it.
Difference tests determine whether a perceptible difference exists between two products, which is useful when ingredients, processes, packaging or suppliers change, or for shelf-life studies. In the triangle test (ISO 4120), each assessor receives three coded samples, two of which are the same, and must identify the odd one; the chance of a correct guess is 1/3. In the duo-trio test (ISO 10399), the assessor receives a reference sample and two coded samples, one of which matches the reference, and must identify the match; the chance probability is 1/2. In the paired comparison test, the assessor receives two samples and is asked which has more of a specified attribute, such as sweetness; this directional test, also called the two-alternative forced choice (2-AFC), has a chance probability of 1/2. In the tetrad test, the assessor receives four samples, two of each product, and must sort them into two groups of two; its chance probability is also 1/3, but it is statistically more powerful than the triangle test for the same number of assessors, because sorting into pairs is a simpler cognitive task. The two-out-of-five test, with a chance probability of 1/10, is efficient but fatiguing. The A-not-A and same-different tests are used when samples cannot be presented together easily.
Tests may be specified (asking about a particular attribute) or unspecified (asking about any difference). Directional tests are more sensitive when the attribute that changes is known, but they assume assessors understand and perceive that attribute.
Results are analysed with the binomial distribution: the number of correct responses is compared with the number expected by chance. For a triangle test with 30 assessors, at a 5 per cent significance level, at least 15 correct responses are needed to conclude that a difference exists; with 12 correct out of 30 the result is not significant. Tables in standards give critical numbers for each test, sample size and significance level, or the exact binomial probability can be calculated. The null hypothesis is that there is no perceptible difference.
A non-significant result does not prove that products are the same; it may reflect too few assessors. When the aim is to show similarity, for example that a cheaper ingredient does not change the product, the test must be designed for similarity: the proportion of discriminators considered acceptable and the β risk (the risk of missing a real difference) are set in advance, and many more assessors are usually needed. Thurstonian models, which express differences as d′ (d-prime), a measure of the size of the sensory difference independent of the test method, allow comparison between tests and explain why the tetrad and 2-AFC need fewer assessors than the triangle and duo-trio for the same difference.
Ranking asks assessors to put a set of samples in order of an attribute or of preference. It is quick, easy for assessors and requires little training, so it suits screening several samples, but it gives only order, not the size of differences: two samples ranked first and second may be almost identical or very different. Ranked data are analysed with non-parametric methods, such as the Friedman test, which tests whether rank sums differ more than expected by chance when each assessor ranks all samples, followed by multiple comparisons of rank sums. Ranking for preference is used with consumers, ranking for intensity with trained or screened assessors. Ranking becomes difficult with more than about five or six samples, especially for strongly flavoured products, because memory and fatigue limit comparisons; balanced incomplete designs, in which each assessor ranks only a subset, are used for larger sets. Ties are usually not allowed, because forcing a choice gives more information, though some procedures permit them. Ranking is also a useful screening step in panel selection, showing whether a candidate can order samples correctly by concentration of a taste or by intensity of an attribute.
Scales measure intensity or liking more finely. Category scales present a limited number of ordered categories, such as a 5-, 7- or 9-point scale, with verbal labels or numbers; they are simple and widely used but may compress responses at the ends and assume equal intervals between categories. Line scales, such as an unstructured 15 cm line anchored with verbal labels near each end, let assessors mark any point, giving continuous data and avoiding favourite numbers; they are the standard in many descriptive methods. Magnitude estimation asks assessors to assign numbers proportional to perceived intensity relative to a reference, so that a sample twice as intense receives twice the number; it produces ratio data suited to psychophysical studies but is harder for untrained assessors. Labelled magnitude scales combine verbal labels placed at empirically determined positions on a line, aiming to give ratio properties with easier use, and include versions for liking.
Scale choice affects data quality. End effects, central tendency (avoiding extremes), contrast effects, halo effects (one attribute influencing ratings of others) and number biases must be considered. Scales should have clear anchors, be appropriate to assessors' training and the number of samples, and be tested before use. Just-about-right scales, which ask whether an attribute is too weak, just about right or too strong, are used with consumers to diagnose formulations, as discussed in chapter 8 and in food-product-development.
Descriptive analysis uses a trained panel to identify and quantify the sensory attributes of products, producing a profile that shows how products differ. The Flavour Profile Method, developed at Arthur D. Little in the late 1940s, uses a small panel of highly trained assessors who reach a consensus profile of a product's aroma and flavour attributes, their intensities on a simple scale, their order of appearance, aftertaste and overall impression (amplitude). The Texture Profile Method, developed at General Foods in the early 1960s by Szczesniak and colleagues, describes mechanical, geometrical and other texture attributes in their order of perception from first bite to swallowing, using standard rating scales anchored by reference foods.
Quantitative Descriptive Analysis (QDA), introduced by Stone and Sidel in 1974, uses a panel of about 10 to 12 assessors who develop their own vocabulary for the product category under a panel leader who does not influence the terms; assessors rate attributes independently on unstructured line scales in replicated sessions, and the data are analysed statistically by analysis of variance, making it suitable for comparing products. The Spectrum method, developed by Civille and colleagues, uses intensively trained panels and universal intensity scales anchored by extensive sets of reference standards, so that intensity ratings can be compared across products and categories.
Classical descriptive methods are powerful but slow and costly, so rapid methods have been developed. Check-all-that-apply (CATA) questions ask consumers or assessors to tick which terms from a list describe a product; they are easy for consumers and, analysed by frequency and correspondence analysis, describe products in consumer language. Free-choice profiling lets each assessor use their own terms, with results combined by generalised Procrustes analysis. Flash profiling asks assessors to rank products on attributes of their own choosing. Projective mapping or napping asks assessors to place products on a sheet so that similar products are close together, giving a map of perceived similarities. Sorting tasks group products by similarity.
The output of descriptive analysis is usually shown as a spider (radar) plot of mean attribute intensities, tables with significance tests, and multivariate maps from principal component analysis, which show which products are similar and which attributes distinguish them, as described in statistics-experimental-design-food. Descriptive data are combined with consumer liking data in preference mapping (chapter 9) to find which attributes drive liking, guiding product development.
Perception changes during eating: sweetness rises and falls, bitterness may linger, texture changes as food is chewed and mixed with saliva, and aroma is released progressively. Static methods give one rating per attribute and miss these dynamics. Time-intensity (TI) analysis asks assessors to record the intensity of one attribute continuously from the moment the sample is placed in the mouth, through chewing and swallowing, to the end of the aftertaste, usually with a computer mouse or slider. Curves are summarised by parameters such as time to maximum intensity, maximum intensity, duration, plateau time and area under the curve. TI is widely used for sweeteners, whose temporal profiles differ from sucrose (some are slow to reach maximum or linger), for bitterness and astringency, for flavour release from chewing gum and for cooling agents. It is demanding for assessors because only one attribute is followed at a time. Assessors need training to use the response device consistently, and individual curves differ in shape, so averaging methods that preserve curve features, rather than simple point-by-point means, are used to summarise panel results.
Temporal Dominance of Sensations (TDS), introduced around 2009, asks assessors to indicate, from a list of attributes displayed on screen, which attribute is dominant, the one catching their attention, at each moment during consumption, and optionally its intensity. The results are shown as dominance curves, the proportion of assessors selecting each attribute over time, compared with chance and significance levels, revealing the sequence of sensations in a product, for example from sweet to fruity to bitter. Temporal check-all-that-apply (TCATA) lets assessors tick all attributes applicable at each moment, capturing several simultaneous sensations.
Temporal methods have been used to study products such as chocolate, wine, coffee, chewing gum and reduced-sugar or reduced-salt products, where reformulation often changes the timing rather than the overall intensity of sensations. They require training on the procedure, carefully chosen attribute lists and suitable software, and their data need specialised analysis, but they provide information on eating experience that static methods cannot.
Affective tests measure liking or preference with consumers who represent the target market. The 9-point hedonic scale, developed by Peryam and Pilgrim for the US Army Quartermaster Food and Container Institute and published in 1957, is the most widely used. Its categories run from 1, 'dislike extremely', through 'dislike very much', 'dislike moderately', 'dislike slightly', 5, 'neither like nor dislike', 'like slightly', 'like moderately', 'like very much' to 9, 'like extremely'. Scores are treated as interval data and analysed by analysis of variance and mean comparisons. Facial (smiley) hedonic scales with fewer points are used with young children, and labelled affective magnitude scales and other scales are alternatives.
Liking can be measured overall and for specific attributes such as appearance, flavour and texture, though attribute liking ratings are often influenced by overall liking (halo effect). Just-about-right (JAR) scales are often added to diagnose why a product is liked or disliked. Penalty analysis combines JAR and overall liking: for each attribute, the mean drop in liking among consumers who found it 'too much' or 'too little', compared with those who found it 'just about right', is weighted by the proportion of consumers in that group, showing which deviations hurt liking most and therefore which attributes to change first.
Preference tests ask which of two or more products a consumer prefers. The paired preference test asks which of two samples is preferred; with no prior expectation about direction, it is analysed as a two-tailed binomial test with chance probability 1/2, sometimes with a 'no preference' option whose handling must be decided in advance. Ranking for preference orders several samples. Preference does not show whether either product is liked: a consumer may prefer one of two disliked products, so hedonic ratings are usually collected as well.
Acceptance and purchase intent questions, food action rating scales, and measures of emotions evoked by products extend affective testing. Context effects also matter: the same product may score differently when tested blind, branded or with price information, and in a laboratory, a hall or at home, so the test conditions should match the decision being made. Results depend on the consumers tested, so recruitment must match the target market in age, gender, region, product usage and other characteristics, and screening questions exclude those who do not use the category or who work in related industries. Affective tests should never be run with trained descriptive panelists, whose training and familiarity with products make them unrepresentative of consumers.
Sensory data are analysed with methods matched to the test and data type. Difference and preference test results, counts of correct responses or choices, are analysed with the binomial distribution, or the chi-square test for larger samples, as described in chapter 4. Ranked data are analysed with non-parametric tests such as the Friedman test for several products ranked by each assessor, and the Wilcoxon or sign tests for two products, as described in statistics-experimental-design-food.
Scaled descriptive data are analysed by analysis of variance. A typical model includes products, assessors and replicates, and the product × assessor interaction; assessors are usually treated as a random effect, because the panel is a sample of possible assessors, so the product effect is tested against the product × assessor interaction rather than the residual error, making conclusions more robust. Significant product effects are followed by multiple comparison procedures, such as Tukey's honestly significant difference or Fisher's least significant difference, to find which products differ. Consumer hedonic data are analysed similarly, with consumers as a random effect. Assumptions of normality and equal variances are checked; with large consumer samples, ANOVA is robust to moderate departures.
Because sensory profiles include many correlated attributes, multivariate methods summarise them. Principal component analysis reduces many attributes to a few components that explain most of the variation, giving maps of products and attributes. Generalised Procrustes analysis aligns individual assessors' configurations in free-choice profiling. Cluster analysis groups consumers with similar liking patterns, revealing segments that would be hidden in average scores: a product with a moderate mean liking may be loved by one segment and disliked by another. Correspondence analysis maps CATA frequencies.
Preference mapping relates liking to sensory attributes. In internal preference mapping, principal component analysis is applied to the consumer liking data, giving a map of products based on liking with consumers as vectors pointing to the products they like, onto which descriptive attributes are then projected. In external preference mapping, each consumer's liking is regressed on the product positions in a sensory map from descriptive data, identifying regions of the sensory space where liking is highest. Partial least squares regression, which links a block of sensory attributes to a block of liking or instrumental data, is widely used to identify drivers of liking. These methods guide reformulation, but their conclusions depend on the range of products tested and should be confirmed with new prototypes.
Consumer research explores and measures consumers' needs, attitudes, habits and responses to products. Qualitative methods explore in depth with small numbers of people. Focus groups bring together about 8 to 12 consumers from the target group for a discussion of one to two hours led by a trained moderator following a discussion guide, often observed by the development team; they generate ideas, language and hypotheses, explore reactions to concepts and prototypes, and uncover reasons behind behaviour. In-depth interviews explore individual views without group influence; ethnographic and in-home observation show how products are actually bought, stored, prepared and eaten, often revealing needs that consumers do not articulate. Online communities and social media analysis extend these methods.
Qualitative findings are not statistically representative: a small group may be dominated by a few voices, and participants say what they think is expected. They are therefore used to generate and refine ideas and questions, which are then tested quantitatively.
Quantitative consumer tests measure responses with enough consumers to support statistical conclusions. Central location tests bring consumers to a controlled venue, such as a sensory facility, shopping centre or hall, where products are prepared and served under standard conditions; they allow many products to be tested with control of preparation and presentation, but in an artificial setting. Home-use tests give consumers products to use in their own homes over days or weeks, under realistic conditions of preparation, use and context, and capture repeated use, convenience and family responses, but with less control. Online surveys measure attitudes, habits, concept appeal and purchase intent in large samples quickly.
Good consumer research requires a clear objective, a defined target population and sampling plan, a sample size large enough for the decisions to be made, well-designed and pre-tested questionnaires that avoid leading or ambiguous questions, balanced presentation orders, and ethical practices: informed consent, protection of personal data, safety checks for allergens and dietary restrictions, and fair treatment of participants. Results should be interpreted with awareness of context effects, since consumers rate products differently in a laboratory, a hall and at home, and of the gap between what consumers say and what they do.
Instrumental measurements are faster, cheaper and more repeatable than sensory panels, so food companies seek instruments that predict sensory attributes. For texture, universal testing machines and texture analysers measure force, deformation and time. Texture profile analysis (TPA), developed from the General Foods texture profile, compresses a sample twice in succession, imitating two bites, and derives parameters such as hardness (peak force of the first compression), cohesiveness (ratio of the work in the second compression to that in the first), springiness, adhesiveness, gumminess and chewiness. Other tests include penetration, shear (for example, Warner-Bratzler shear force for meat tenderness), three-point bending (for biscuit snap), extrusion and rheological measurements of viscosity. Acoustic recording of fracture sounds supports crispness measurement. Rheology is covered in food-physical-properties-rheology.
Colour is measured by colorimeters and spectrophotometers in the CIE L*a*b* system, in which L* is lightness, a* the red-green axis and b* the yellow-blue axis. The total colour difference between two samples is ΔE = √(ΔL*² + Δa*² + Δb*²); for example, differences of 2 in L*, −2 in a* and 1 in b* give ΔE = 3.0. Differences above about 2 to 3 are commonly visible to observers, though perceptibility depends on the colour region and conditions, and more uniform difference formulae have been developed.
For aroma, gas chromatography-mass spectrometry identifies and quantifies volatile compounds, and gas chromatography-olfactometry, in which a trained assessor sniffs the column effluent and records odours, identifies which of the many volatiles are odour-active. Odour activity values, the concentration of a compound divided by its odour threshold, estimate which compounds contribute to aroma. Electronic noses and tongues, arrays of chemical sensors with pattern recognition, discriminate samples and are used in quality control, but they do not replicate human perception and need calibration against sensory data.
Instrumental and sensory data are linked by correlation and regression, including partial least squares, as described in statistics-experimental-design-food. Strong correlations allow instruments to replace some routine sensory tests in quality control, for example hardness by texture analysis or colour by colorimetry. But correlations hold only over the range of products studied, perception integrates many physical properties non-linearly, and instruments cannot measure liking. Sensory evaluation therefore remains the reference, and instruments are validated against it. A practical sequence is to build a calibration set of products spanning the expected range, measure them by both panel and instrument, fit and validate a model on separate samples, and then recheck the relationship periodically, especially when raw materials or processes change.
Judging and grading are the traditional sensory assessment of quality, carried out by experienced judges or graders against defined standards, as distinct from research sensory testing. Score cards allocate points to quality characteristics, such as flavour, body and texture, colour and appearance, and package, with the most points for the most important characteristic, usually flavour, and list typical defects and their deductions. Score cards have long been used in dairy products, for example for milk, butter, ghee, cheese and ice cream in dairy science courses and competitions; in meat, fish, tea, coffee, wine and spices; and in bakery judging. Graders of commodities such as tea, coffee, wine, cocoa and spices taste many samples daily against reference standards to set prices and blends.
Official grading schemes combine sensory and analytical criteria. In India, AGMARK grade standards for products such as ghee, butter, honey, spices and pulses include sensory requirements together with chemical limits, as described in food-law-regulatory-affairs; international standards for commodities such as olive oil include sensory assessment by trained panels, which classify virgin olive oils by the presence and intensity of defects and the fruitiness attribute.
Expert judging is fast and draws on deep experience of a product category, defects and market requirements, and remains essential in commodity trading, quality control and competitions. But it has limits. Judges' scores combine quality and personal preference, are often not validated statistically, may not reflect consumer liking, and can vary between judges and over time. Some score cards weight attributes in ways that do not match consumer importance. Modern quality systems therefore use trained panels with defined attributes and statistical monitoring for routine quality control, often with a quality rating relative to a standard (in, out or degree of difference from a gold standard), and link quality specifications to consumer research so that 'quality' reflects what consumers value.
For routine factory quality control, sensory checks are built into specifications: each batch or shift is tasted and examined by trained staff against reference samples, with defined acceptance criteria and procedures for holding or rejecting product. Records are kept, and panel performance is monitored as for research panels. The sensory part of quality control catches problems, such as off-flavours from contaminated ingredients or packaging taints, that chemical and microbiological tests may miss.
A student uses this subject to learn which test answers which question and how to analyse it: triangle or tetrad for difference, QDA or CATA for description, the 9-point hedonic scale for liking, and the binomial, ANOVA and Friedman tests for the results. Running a triangle test in class, calculating whether the number of correct responses is significant, and setting up booths with coded samples and balanced orders fixes the principles.
Five years into a food company, a sensory scientist or technologist runs difference tests for supplier and process changes, maintains a trained descriptive panel, organises consumer tests for new products, analyses data and presents clear recommendations, sets up sensory checks in factory quality control, and links sensory results to instrumental measurements and product specifications.
A senior sensory and consumer insight leader with twenty-five years' experience designs the company's sensory programme and facilities, chooses methods and suppliers, ensures that decisions on reformulation, cost reduction and launches rest on valid sensory and consumer evidence, builds global consumer databases and benchmarks, and trains product developers to ask the right sensory questions.
A researcher at this level studies perception and its measurement: individual differences in taste and smell, cross-modal interactions, temporal and emotional measures, rapid descriptive methods, Thurstonian modelling, the validity of consumer tests in context, and links between sensory perception, food structure and health-related reformulation. Such work combines physiology, psychology, statistics and food science.
This manual explains methods and analysis in principle; it does not reproduce critical value tables for difference tests, full score cards for each product, or software procedures. Thurstonian modelling, sample-size calculation for similarity tests and advanced multivariate methods are introduced only briefly. Values such as panel sizes and colour difference thresholds are typical ranges that vary with purpose and standard. Product development applications are in food-product-development and statistical methods in statistics-experimental-design-food.
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.