The EnoMatch method:
Sensory model and study design

What we want to investigate

Our working hypothesis is that a smaller distance between a consumer’s Enotype and a wine’s Winetype is associated with greater preference for that wine. The EnoMatch study aims to assess whether this relationship offers useful predictive information beyond pre-specified, non-personalized recommendations.

We are not seeking to identify a universally ideal wine or establish a hierarchy of palates. We want to test a specific method, report where it works and where it does not, and distinguish numerical similarity from demonstrated preference. Placing consumers and wines on axes ranging from 0 to 100 does not, by itself, prove that those coordinates predict liking.

Model described: the Q20r3 revision, with two-dimensional Structure/Balance coordinates and five profile regions. Earlier versions used different questions, classifications or calculations; their historical results must remain identifiable by version.

The mathematical model

EnoMatch uses two coordinates for each consumer and each wine: Structure (S) on the horizontal axis and Balance (B) on the vertical axis. Both run from 0 to 100. The public notation is S/B (for example, 54/67), where the slash separates two coordinates rather than indicating a fraction.

Enotype: the consumer’s preference profile

Enotype is a two-coordinate estimate of a consumer’s sensory preferences, derived from questionnaire answers. The current questionnaire contains 20 main questions: 15 contribute to its calculations and five are exploratory. A separate five-choice calibration question, Q21, is answered before the result is shown but does not influence the Enotype coordinates.

The scored questions produce four internally weighted preference indices, each on a 0–100 scale. They relate to TN (tannic/astringent sensations), AL (alcohol-related sensation), AC (acidity) and SW (sweetness). The five exploratory questions are Q1, Q3, Q10, Q11 and Q13; biological sex is recorded by Q1 but is not used as a scoring factor.

The consumer’s coordinates are calculated as follows:

The subscript \(E\) denotes the consumer. The weights are model assumptions that must be tested, not experimentally established contributions to wine preference. The consumer’s Balance equation is linear and has no additional squaring or exponentiation step.

For example, when all four preference indices equal 50, the resulting Enotype is 50/50, classified as Neutro under the current five-profile system. Internally the coordinates retain their full precision; rounding is used only for display.

Enotype summarizes self-reported preferences. It is not a direct measurement of genes, taste-bud density, or clinical sensory sensitivity; nor does it assess personality, leadership or sociability. A profile estimated from an in-person tasting is a different, exploratory procedure and cannot be assumed equivalent to the online questionnaire without independent calibration.

Winetype: the wine’s sensory profile

Winetype describes a wine using two calculated coordinates, based on sensory ratings and analytical or technical information. It uses five inputs: tannicity, acidity, alcohol content, CO₂ pressure and residual sugar. The numerical equations for wine differ partly from those for consumers.

The tannicity and acidity inputs are sensory judgments, not automatic conversions from tannin concentration, pH or titratable acidity. Their reference scales, assessor training and reproducibility need to be specified and evaluated. The boundaries above are computational normalization limits, not claims that all wines fall in those analytical ranges. Missing input data must not be silently replaced with zero.

Normalizing alcohol, pressure and sugar

For the calculation, the three measured inputs are constrained to predefined ranges:

The alcohol-related index incorporates a model-defined CO₂ correction:

For a wine with no CO₂ pressure in the calculation, 8.5% and 15.5% vol. correspond to index values of 0 and 100. At 6 bar the pressure factor is 0.9: it reduces the alcohol-related component by 10% relative to the same alcohol content at 0 bar, not the whole Winetype. This adjustment is a model assumption, not an established physiological law.

The sugar-related index follows a logarithmic transformation:

Under this model, 0 g/L maps to 0, 20 g/L to 50, and 200 g/L to 100. Sugar concentration is not identical to perceived sweetness, which also depends on acidity and other factors.

Calculating the wine’s two coordinates

Winetype Structure combines the wine’s sensory tannicity and alcohol-related index:

Winetype Balance first calculates an intermediate value:

That intermediate value is transformed using the fixed exponent \(\gamma=0.7\):

Here \(\operatorname{sgn}\) is the sign function. The transformation preserves the 0, 50 and 100 reference points. This is not the same Balance transformation used by Enotype; consequently, equal consumer and wine input indices do not necessarily produce equal Balance coordinates. That asymmetry is explicitly part of the current model and needs empirical testing.

The five EnoMatch profiles

Enotype and Winetype are placed on the same S/B map, with Structure (S) increasing left to right and Balance (B) becoming more tart / acidic from bottom to top. The horizontal and vertical axes both run from 0 to 100. A lower Balance value is described as more mellow; it does not necessarily mean analytically sweet wine.

The central Neutro region is a square spanning 35–65 on both axes, inclusive. It takes priority over the four outer regions. Classification is based on unrounded coordinates, not the rounded S/B label displayed to the public.

For example, 50/50 is Neutral, not Smooth. Points exactly on S = 35 or 65 and B = 35 or 65 are inside Neutral when both coordinates satisfy the central rule. Profiles are interpretative labels, not discrete biological types, measures of wine quality or guarantees of enjoyment. The earlier designation Round was replaced by Denso in this model revision; historical records retain their original version and labels.

Affinity: comparing Enotype and Winetype

The EnoMatch Affinity index is based on the weighted distance between the two-dimensional consumer and wine profiles:

All calculations use unrounded coordinates. The 60/40 weights apply to the squared coordinate differences inside the distance formula. A perfect numerical match gives 100; across the defined 0–100 axes, the maximum possible distance is 100.

Worked example — not a study result. Imagine a consumer with Enotype 54/67 and a hypothetical wine with Winetype 60/70:

This example explains the geometry only; it does not show that the consumer would like that wine. Affinity is a similarity measure, not a calibrated probability of liking.

Q21: an independent calibration question

Before receiving the test result, participants select one of five broad wine-style descriptions. The answer is stored independently: it does not affect the Enotype coordinates or profile classification. Each answer corresponds to an abstract reference point on the map:

Each reference point is a geometric convention, not a measured wine. We calculate the distance between the consumer’s Enotype and the selected point using the same 60/40 distance formula. This produces Affinity_Q21, which is an exploratory measure of agreement with a stated style, not evidence of real-wine preference.

How the study is designed

1. Online phase: preferences and calibration

Participants answer 20 questions, provide the demographic and participation information specified in the consent materials, and then select one of the five Q21 wine-style descriptions before seeing their Enotype. Q21 can be compared with their calculated position but does not feed into it.

We plan to examine results across consumer groups, including wine professionals and other consumers. The study will keep track of questionnaire version, language, missing data, and the wording of Q21. Spanish and English versions use corresponding question identifiers and scoring rules; numerical equivalence does not automatically prove equivalent comprehension across languages or cultural settings.

2. In-person tasting: observed wine preferences

The tasting component tests whether Enotype-derived recommendations correspond to what people actually like when wines are served without brand or price cues. Each wine’s Winetype must be determined in advance using a documented characterization protocol, not adjusted after observing the participants’ preferences.

Our proposed controls include coded samples, a balanced or randomized serving order, consistent temperature and serving conditions, individual liking ratings before discussion, and documentation of prior exposure to a participant’s Enotype. Expectations can influence perceived pleasantness, which is one reason to conceal commercial information during tasting [6].

Preliminary in-person observations may help refine the protocol. Such development data are not equivalent to an independent confirmatory validation. In particular, a physical profile constructed by averaging wines’ coordinates weighted by the participant’s own liking ratings must not be treated as a separately measured true Enotype. Evaluating it against those same ratings can introduce circularity.

3. Pre-specified analysis and transparent reporting

The report should document recruitment, sample size, wine selection, Winetype characterization, survey and tasting procedures, exclusions, statistical methods and uncertainty. It should include negative as well as positive findings and distinguish exploratory from confirmatory analyses.

Data used to revise questions, weights, profile boundaries or formulas are development data. An updated model should be frozen and assessed on independent participants. Selecting and evaluating a model on the same observations can overstate its performance [7].

We intend to register the confirmatory study protocol before the relevant new data are collected. This page is not a completed preregistration; a reference to the public registration will be added when one exists [8].

How we will evaluate performance

The evaluation must distinguish agreement with a declared style (Q21) from predictive performance for actual wines tasted. The main comparison concerns the wines offered in the study, not every bottle on the market.

Mean weighted S/B distance to the favorite wine

For participant \(i\), let \(f(i)\) denote the wine identified as their favorite in the blind tasting. Its measured Winetype has coordinates \((S_{W,f(i)},B_{W,f(i)})\). A proposed distance-based outcome is:

Lower distance means that the Enotype coordinates are closer, on average, to the Winetype of the participant’s selected favorite. It is a mean weighted distance, not the old one-dimensional mean absolute error (MAE). The mean Affinity to the favorite is \(100-\overline d_{\mathrm{favorite}}\), a restatement of this outcome rather than independent evidence.

A constant 50/50 Enotype provides one simple baseline, but the study must also compare other pre-specified, non-personalized recommendations. A good numerical distance alone cannot demonstrate that the questionnaire is personalized or causally predicts liking.

Top-choice accuracy

A complementary outcome asks whether the wine recommended by the highest Affinity is the participant’s favorite among the tested wines:

Here \(k\) counts participants for whom the model’s highest-ranked wine matches their stated favorite, and \(n\) is the number of valid participants. Tie rules and treatment of incomplete ratings must be specified in advance. This should be compared with a non-personalized recommendation rule applied to the same wines.

Liking ratings and ranking quality

We will also consider the liking score assigned to the recommended wine, whether it is among a participant’s top-ranked wines, and the relationship between Affinity and their ratings across wines. A wine that is not the first choice may still be well liked. These analyses should account for repeated observations from the same participant.

Uncertainty and external validity

Findings will be presented with sample sizes and appropriate uncertainty estimates. A small pilot or one set of four wines cannot establish generalization to all consumers, wine regions, styles or occasions. Any performance statement will be limited to the samples, wine characterization and version actually tested.

What the model does not capture

Wine is a multisensory experience involving smell, taste, mouthfeel, expectations and context [3]. Two S/B coordinates inevitably simplify this complexity.

Structure is an index derived from tannic and alcohol-related components, not a complete physical measurement of body. Balance models the relationship between acid-related and sweetness-related components, not a direct analytical classification as dry or sweet. Wines with very different aroma, texture or composition may share similar S/B coordinates.

Four underlying indices (TN, AL, AC and SW) can be visualized separately for research, but the current public Enotype and Winetype are two-coordinate profiles, not validated four-dimensional sensory maps. The overlapping areas of radar charts do not constitute a validated matching measure, and their geometry can depend on axis order [11].

Differences in perceived acidity or tannicity between tasters, variation in wine measurements and the distinct consumer/wine Balance transformations may affect observed performance. These are issues to assess experimentally, not reasons to assume the model is already reliable.

Taking part

Adults aged 18 or over who already consume wine may participate in the online questionnaire. Both consumers and wine professionals are welcome. No wine tasting products or specialist knowledge are needed to answer the online questions. Participation is voluntary and subject to the participant information and privacy notice.

What will I receive?

When the current five-profile questionnaire is available, completing the test shows your Enotype as S/B, your position on the map, the assigned profile (Light, Bold, Smooth, Rich or Neutral) and a descriptive interpretation. The test is designed to send a copy of the result by email. These descriptions are guides, not diagnoses or guarantees of enjoyment.

Does completing the questionnaire register me for an in-person tasting?

No. The online questionnaire and any in-person study session are separate activities. Participation conditions for tasting sessions are communicated separately.

What information does the form request?

The form asks for a name, email address, country, year of birth, participant type and answers to the questionnaire. Wine professionals provide relevant professional details. Biological sex is an exploratory, non-scored question, and the current questionnaire also includes other non-scored preference questions. Please review the participant information and privacy notice before submitting your responses.

Where will the findings be available?

We intend to publish the research report here, including methodological limitations and any findings that do not support the hypothesis. Receiving a personal Enotype result does not mean the research has been completed or scientifically validated.

Take the Enotype test

The team behind EnoMatch

EnoMatch is developed by two co-authors responsible for the project’s direction and development. The people developing the method also promote the study; this relationship will be disclosed in research reporting.

Jordi Rustarazo
Co-author · Sommelier & Economist

Miguel Ángel Gil
Co-author · Engineer & MBA

For questions about the project or participation: info@enomatch.com.

Project status and future findings

The method is in development and evaluation. The Q20r3 model adds a central Neutral profile and replaces the earlier Round label with Rich, while keeping its two-axis numeric equations. Its five-choice calibration question changes the theoretical Q21 baseline, but does not itself demonstrate better predictive performance.

No completed independent validation report for the current EnoMatch model is presented on this page. Any final report should identify the exact model version, disclose development analyses separately from independent validation, and describe the extent of the evidence. Earlier versions, exploratory tastings and questionnaire data must not be merged without accounting for changes in scoring and classification.

Scientific background and references

Sensory research shows that perceived intensity and liking are distinct and that wine preference is affected by individual oral responsiveness, saliva, mouthfeel and expectations [1] [2] [4] [5]. These publications motivate the research question; they do not validate EnoMatch or imply that their authors endorse or participate in the project.

Origins and scope

The project’s initial thinking drew on research by Borchgrevink and Sherwin concerning wine preferences and the work of Tim Hanni. EnoMatch uses its own adapted questionnaire, two-dimensional wine and consumer indices, profile classification and matching rule. It does not claim equivalence to other similar classifications or psychological personality types.

Sensory science and wine preference

[1] Hayes, J. E., & Keast, R. S. J. (2011). Two decades of supertasting: where do we stand? Physiology & Behavior, 104(5), 1072–1074. doi:10.1016/j.physbeh.2011.08.003. Individual sensory differences and limits on inferring preferences from responsiveness.

[2] Cravero, M. C., et al. (2020). Profiling Individual Differences in Alcoholic Beverage Preference and Consumption: New Insights from a Large-Scale Study. Foods, 9(8), 1131. doi:10.3390/foods9081131. Beverage preferences and measured oral responsiveness in an Italian sample.

[3] Shepherd, G. M. (2015). Neuroenology: how the brain creates the taste of wine. Flavour, 4, 19. doi:10.1186/s13411-014-0030-9. Multisensory aspects of wine experience.

[4] Niimi, J., Danner, L., Li, L., Bossan, H., & Bastian, S. E. P. (2017). Wine consumers’ subjective responses to wine mouthfeel and understanding of wine body. Food Research International, 99, 115–122. doi:10.1016/j.foodres.2017.05.015. The impact of mouthfeel manipulations on consumer responses.

[5] Luo, J., et al. (2023). Variation of wine preference amongst consumers is influenced by the composition of salivary proteins. npj Science of Food, 7, 51. doi:10.1038/s41538-023-00222-1. Exploratory research on saliva and preference.

[6] Plassmann, H., O’Doherty, J., Shiv, B., & Rangel, A. (2008). Marketing actions can modulate neural representations of experienced pleasantness. Proceedings of the National Academy of Sciences, 105(3), 1050–1054. doi:10.1073/pnas.0706929105. The relevance of expectations to tasting conditions.

Research methods

[7] Cawley, G. C., & Talbot, N. L. C. (2010). On Over-fitting in Model Selection and Subsequent Selection Bias in Performance Evaluation. Journal of Machine Learning Research, 11, 2079–2107. Read the paper.

[8] Center for Open Science. Preregistration. Methodological resource. A resource for prospective study registration, not an EnoMatch registration record.

[11] Heijungs, R. (2022). Two arguments against the use of radar plots for constructing composite indicators. Brazilian Journal of Chemical Engineering, 39, 885–886. doi:10.1007/s43153-022-00247-1. Cautions about radar-chart area-based composites.

Background to personalized wine recommendation

[9] Hanni, T. (2013). Why You Like the Wines You Like: Changing the Way the World Thinks about Wine. HanniCo. Book information. A conceptual antecedent, not validation of EnoMatch.

[10] Borchgrevink, C. P., & Sherwin, A. L. (2017). Predicting wine preference: testing the premises of the vinotype theory. International Journal of Wine Business Research, 29(3), 251–268. doi:10.1108/IJWBR-08-2016-0027. Research concerning food and beverage preferences as predictors of wine preference.