Why Smell Has No Primary Colors
Updated: Oct 20
The idea is irresistible: if color has red, green and blue, why should smell not have a few basic odors that combine into everything else?
It would make fragrance wonderfully tidy. Learn the primaries, mix them in the right proportions, and every scent could sit somewhere on one universal wheel.
That is not how olfaction is currently understood.
Science has not established a small RGB-like set of primary odors that can be mixed to predict ordinary smells. That is narrower than saying smell has no structure, no dimensions or no useful models. It says the color analogy stops being predictive much earlier than we might expect. [2]
What “primary” would have to mean
For the analogy to do real scientific work, a small set of odor primaries would need to be more than familiar categories such as floral, woody or citrus.
A useful primary system would need to provide a compact basis from which a broad range of ordinary odor percepts could be reconstructed or predicted by reliable combination rules.
That is the attraction of the color comparison. In additive-light systems such as displays, red, green and blue primaries are engineered to stimulate the three cone classes in combinations that reproduce a wide gamut of perceived colors.
That does not mean humans “see only three colors,” and it is not a lesson about mixing paint. RGB is an additive-light model built around the response properties of the visual system.
Smell has a different starting architecture.
The first break in the analogy: receptor coding
Olfaction does not assign one receptor to one familiar smell.
Classic receptor work by Malnic and colleagues showed a many-to-many pattern: an odorant can activate multiple receptor types, and an individual receptor can respond to multiple odorants. Odor identity is represented by combinations of receptor activity rather than a one-receptor/one-note dictionary. [1]
That matters because a consumer word such as “rose,” “cedar” or “vanilla” is not a label for one receptor channel.
The receptor code can also change with molecular structure and concentration. So even before perception is assembled into a recognizable odor, the input is already a distributed pattern rather than three or four obvious axes. [1]
Receptor space is not perceptual space
A second distinction is just as important: a receptor-response pattern is not the same thing as the odor you consciously experience.
Researchers can describe activity across receptors, neural populations, molecular features or learned mathematical dimensions. A person, meanwhile, reports percepts such as smoky, green, floral, sharp, familiar or hard to name.
Those are different levels of description.
Barnum and Hong’s modern review emphasizes that olfaction does not come with an obvious ordered physical axis comparable to wavelength in vision, and that finding the most useful dimensions of odor space remains an active research problem. [2]
So hundreds of receptor channels do not imply that conscious smell must have hundreds of equally important perceptual dimensions. The reverse is also true: finding a compact perceptual representation would not prove that its axes are literal biological “primary smells.”
Keep three spaces separate
Physical or molecular space: properties of odorant molecules and mixtures.
Receptor/neural space: patterns of biological response to those odorants.
Perceptual space: the similarities, qualities and dimensions people experience and report.
Models can connect those spaces, but they should not be collapsed into one color wheel.
Mixtures add another problem
Even if single odorants could be arranged neatly, mixtures would still have to obey predictable combination rules for a true RGB-like system to work broadly.
They often do not.
In controlled human work, tested binary odor mixtures have shown intensity hypo-additivity rather than simple arithmetic addition. [3] In a separate defined system, receptor antagonism was associated with a specific suppression effect in human odor perception. [4]
Those studies do not prove that every odor mixture is nonlinear in the same way. They show why “add the components and read off the result” is not a dependable universal rule.
The tested examples show departures from simple linear addition at perceptual intensity and, in one defined case, at receptor/sensory interaction levels. That is enough to show why a few universal sliders cannot be assumed. [3][4]
This is not the same argument as saying complexity is random. It is an argument against assuming linearity.
What about odor maps and AI models?
The absence of accepted RGB-like primaries does not mean researchers have given up on organizing odor.
Quite the opposite.
Modern work builds perceptual spaces, statistical embeddings and learned representations that try to predict relationships between molecules and odor descriptions. Lee and colleagues’ principal odor map, for example, learned a molecular representation that generalized across several odor-prediction tasks. [5]
That is meaningful progress. It shows that odor structure contains learnable regularities.
But a successful representation is not automatically a set of perceptual primaries.
A model can be useful because it compresses information, predicts labels or organizes similarity. Its dimensions do not have to correspond to a few independently mixable smells that generate ordinary odor experience the way RGB channels generate colors on a display. [2][5]
This is the key uncertainty to preserve: useful low-dimensional representations may exist for particular tasks or restricted odor spaces, while a universal small primary set for ordinary odor prediction remains unestablished.
What this changes for a fragrance reader
This science makes several everyday fragrance conventions easier to interpret.
Scent families are useful descriptive maps, not receptor primaries.
Fragrance notes are communication tools, not coordinates in a universal biological color space.
A “wheel” can help people compare language and impressions without claiming that neighboring sectors are fundamental dimensions of the olfactory system.
For the descriptive side of that language, see Fragrance Notes 101: Top, Middle, Base Explained and Scent Families: Citrus, Woods, Amber, Tea—A Calm Guide.
That distinction lets us keep the useful vocabulary without pretending the vocabulary is the sensory code itself.
Why this does not let us reverse-engineer a candle
Receptor-coding research does not tell us the exact formula of a fragrance from its smell, and a learned odor map does not turn a candle into a few primary coordinates.
A finished candle brings together its material composition, release into the air, mixture interactions, room conditions and an individual observer. Receptor-coding research illuminates one part of that process; it cannot, by itself, predict the complete scent experience.
The practical benefit is conceptual rather than predictive: there is no scientifically established tiny palette of basic smells that lets a shopper or maker reconstruct ordinary fragrances by simple mixing.
Reader takeaway
Color is a useful analogy because it shows what a compact sensory model can look like.
Smell is useful precisely because the analogy breaks.
Odorants engage combinations of receptors. Receptor activity and perceptual experience are different spaces. Mixtures do not obey one universal linear rule. Researchers can still build useful representations and discover lower-dimensional structure without having found a small set of universal primary smells.
The safest conclusion is therefore not “smell has no structure.” It is: olfaction has no accepted small RGB-like primary system that predicts ordinary odors by simple mixing.
Comments