What a foundation is actually moving

A foundation is sold as a colour. There is a number on the bottle, a name for the number, and a small ritual at the counter for finding yours. What the perception literature has measured about facial skin is a different quantity — not the average colour of it, but how much that colour varies from one patch to the next — and the two have been separated cleanly enough that it is worth knowing which one you are buying.

The separation was done by taking the shape out. Fink, Grammer and Matts built stimulus faces standardised for three-dimensional shape and for surface topography, so that the only thing left differing between them was the distribution of colour across the skin — the distribution that arrives with biological age and accumulated sun. Faces carrying the even distribution of young skin were judged younger, healthier and more attractive than faces carrying the uneven distribution of older skin. No wrinkle moved. No feature moved. Fink and Matts restated the design plainly two years later: colour distribution influences perceived age and attractiveness independent of surface topography cues.

Then it was measured on real skin rather than composites. Matts, Fink, Grammer and Burquest cropped cheek images from 170 girls and women between eleven and seventy-six and had 353 people rate them for attractiveness, health, youth and apparent age. Homogeneity of the image correlated above .40 with all three of the positive judgements and −.45 with estimated age. They also split each image into melanin and haemoglobin maps and measured the evenness of those separately: the haemoglobin map's homogeneity tracked the raw image's at .92, the melanin map's at .68, and both ran negative against estimated age. The unevenness a stranger is reading is the unevenness of the two pigments underneath.

It is not a local taste. Fink and colleagues took the same kind of cheek images from British women to the Maasai in Tanzania and the Tsimane' in Bolivia — observers with no everyday exposure to lightly pigmented skin at all. Both groups picked the homogeneous images as younger and healthier, and preferred them.

The sensitivity is fine, too. Samson, Fink and Matts took one 61-year-old woman's photograph, removed surface topography and smoothed colour unevenness in 25% steps, and showed 160 German observers every pairing. A quarter of the unevenness gone was already enough to move which face was called healthier. But the two cues divide the work between them, and the division is the part that matters here: topography carried age more strongly, colour distribution carried health more strongly. A cream can flatten a distribution. It cannot fill a line.

Which is roughly what the industry's own instruments say a foundation is. A L'Oréal group working with RIKEN published a hyperspectral method in 2024 that opens by conceding the problem: coverage, they write, is a perceived attribute not captured by opacity or any other single optical property. What they built instead was two numbers — a homogeneity factor for how far the spectra across a face converge after application, and a spectral shift factor for how far the colour moves. The homogeneity factor tracked trained rankings of coverage. Their own caveat is the honest part: it does not fully describe coverage either, and how much a product shifts a given face depends on the difference between that face and the shade. Two jobs. Reduce a variance, move a mean. Only one of them is the job twenty years of perception studies measured.

The same company also measured what a foundation does across a working day and published a result no advertisement would run. Two hundred and seventy Japanese women aged thirty to fifty-four photographed themselves before applying their own foundation in their own way, immediately after, and five hours later — 810 selfies, graded by algorithm. Several of the less-marked ageing signs improved. Periorbital wrinkles were graded more severe at both times, in every age group, and nasolabial folds at five hours. The amplitudes were small, decimals of a grade. They were also significant, and they pointed the wrong way.

Even the ladder the shades are arranged on is a choice somebody made. When L'Oréal measured skin colour across French and American Caucasian, Japanese, African-American and Hispanic-American women in order to build a worldwide foundation range, what came back was a large, continuous, overlapping colour space with light, medium and dark distinguishable inside every group. Only the Japanese women in that study declined the lightness scale, and said a pink–ochre–beige axis made more sense for their skin. Depth survived as the organising axis because it is easy to shelve, not because it is the only axis in the data.

And the experiment the whole ritual assumes has not been published. Search the biomedical indexes for a controlled comparison between a correctly matched shade and a near miss, judged by observers, and nothing comes back. What does come back is odd enough to be worth the detour: papers in dentistry, where a commercial foundation shade guide is used as the instrument for classifying a patient's complexion before a tooth shade is chosen. The shade ladder is trusted enough to measure people with, and has never been tested at the job it was built for. Matching to the jawline is craft. It may well be good craft. It is not a result.

One thing about how much has been tested, and it is not comfortable. Jones, Kramer and Ward let observers dial cosmetics up and down on a series of faces — once for their own taste, once for what they believed women preferred, once for what they believed men preferred. Men and women agreed with each other about the amount they liked, overestimated what women wanted, and overestimated what men wanted by more. The models' own self-applied makeup sat well above every one of those settings.

None of which says the shade is nothing. A mean that is wrong enough is visible from across a room. It says that the shade is the half of the purchase with no evidence under it and the evenness is the half with two decades of it, and that a shelf is arranged the other way round. Percol looks at colours next to a face and hands back an opinion. This is one of the places where the opinion is worth less than the measurement, and the useful thing is knowing which you have been handed.