← Blog

Right Brain · Reading

Why People Saw The Dress Differently

In 2015 a badly lit photograph of a dress divided the internet into two camps who could not believe the other camp was being sincere. Nobody was lying, and there was no filter. What the picture caught was a step your visual system performs on every scene you have ever looked at — a step that normally succeeds so completely you have no idea it is happening. Take in the whole. Before the parts.

By ComunAI · 3 August 2026 · ~6 min read

Your eyes do not measure colour

Start with the problem the visual system is actually solving. The light arriving at your eye from any surface is the product of two things multiplied together: what the surface reflects, and what the illumination happens to be. A white page under warm evening light sends your eye a physically orange signal. A yellow page under cool blue light can send a similar one.

If you reported the signal, the page would change colour all day. It doesn't. The visual system discounts the illumination and reports the surface — an achievement called colour constancy, and it is one of the better-established facts in vision science.¹ Strong Edwin Land's demonstrations made the point unforgettable: he could change the illumination on an array of coloured patches dramatically, and observers kept naming the patches correctly.² Strong

Subtracting the light requires guessing what it was

To remove the illumination from the signal, the system needs an estimate of the illumination. Normally the scene is full of clues — a broad view, familiar objects, shadows, highlights, a range of surfaces to compare against. The estimate is made below awareness, it is usually right, and you never notice it was made.

The dress photograph removed almost all of those clues. It was overexposed, tightly cropped, with no reliable reference surface and no visible light source. That left the illumination genuinely underdetermined — and a system that must subtract something is then forced to assume.

Two assumptions were available, and they lead to opposite conclusions from identical pixels:

• Assume the dress is in shadow / cool bluish light. Subtract blue, and what's left is white and gold.
• Assume the dress is under warm artificial light. Subtract yellow, and what's left is blue and black.

Both are competent inferences. They just start from different priors about the light. The dress was, physically, blue and black — but "which is correct" is much less interesting than the fact that two visual systems, given the same input, produced stable and confidently different objects.

What the vision scientists found

The research response was unusually fast — three papers appeared together in Current Biology within months.³ The core findings held up well:

• The split was real, and it was roughly stable within a person. Rosa Lafer-Sousa, Katherine Hermann and Bevil Conway surveyed a large sample and found the population divided into distinct groups, with individual reports largely consistent rather than random.⁴ Strong A minority also reported a third percept, and some observers could flip between them.
• The blue–yellow axis is special. Alissa Winkler, Lothar Spillmann, John Werner and Michael Webster showed the ambiguity is not symmetric across colour space: the confusion sits along precisely the axis that natural daylight itself varies along, from blue sky to warm sun.⁵ Moderate That is a satisfying detail — the illusion exploits the one dimension your system has the most reason to discount.

The best-supported explanation, then, is about illumination priors: differing assumptions about the light, applied automatically. Moderate

The part that is not settled

The obvious next question — why do people differ in their priors? — has a popular answer that deserves a much weaker grade than it usually gets.

Pascal Wallisch proposed and tested the idea that lifelong exposure matters: people who spend more waking time in daylight might be readier to assume bluish natural light (seeing white and gold), while people who spend more of it under artificial light assume the opposite. He reported an association with self-reported chronotype in a large online sample.⁶ Early

It is an elegant hypothesis and it is not established. The study is correlational, self-reported and online; the effect is a modest tendency, not a rule; and no one has demonstrated the causal path. Treat "night owls see it differently" as an interesting proposal, not a fact about you — and be sceptical of any quiz claiming to read your sleep habits off which colours you named.

Why one photograph mattered so much

The dress became famous for the wrong reason — as a novelty, or a test of whose eyes were "right". What it actually did was make a normally invisible process briefly visible to millions of people at once.

The genuinely useful takeaway is not about colour. It is that seeing is a construction, built from an ambiguous signal plus a set of assumptions, and that your assumptions are not universal. That is the same lesson as the checker-shadow illusion and the rest of the family we went through in Why Optical Illusions Fool Everyone — and it survives knowing about it. Learning the physics does not let you see the other version at will, which tells you the process runs well below the level thinking can reach.

There is a mild social corollary worth stating, given how the argument went in 2015. Someone reporting a different percept from yours, on an ambiguous stimulus, is not being contrary. They are running the same machinery with a different prior, and neither of you has access to the raw data to settle it.

What this means for Right Brain

Right Brain is named for a metaphor, not a hemisphere — the open, whole-first way of looking, as against the narrow, part-by-part one. The dress is the ideal advertisement for the whole-first half of that: colour is not a property read off a patch in isolation, it is decided by the context around the patch — the surrounding surfaces, the inferred light, the whole scene. Crop the context away and the answer becomes undecidable. The colour and illusion games in the app are that idea shrunk to a few quiet minutes: no score, no verdict about your abilities, just your own inference machinery caught briefly in the act.

And the standing limit applies as always: there is no good evidence that playing perception games improves your everyday seeing. Broad transfer is the weak link in this whole field, and we covered it properly in Do Brain-Training Games Actually Work? Contested