"The Dress" (Blue/Black vs White/Gold)

One photo, two dresses

A photograph of a striped lace dress, taken in poor, washed-out light, split viewers into two camps. Some see white fabric with gold lace; others see blue fabric with black lace. Each group is equally sure, and many people can't see the other version even when told about it. The real dress, sold by the British retailer Roman Originals, is royal blue and black. The pixels in the photo are neither: they're a washed-out light blue and a brownish gold, which is why both readings are possible.

The photograph belongs to Cecilia Bleasdale, so it isn't reproduced here; you can see it on Wikipedia. The demo above isolates the part that matters: the same two stripe colours, placed in two different lighting contexts.

Guessing the light

The explanation vision scientists settled on is colour constancy, the same process behind the checker shadow illusion. The brain doesn't simply register the colour of the light reaching your eye; it estimates the colour of the light falling on the scene and discounts it, to work out what colour the surface really is.

The dress photo is unusually bad at revealing its lighting. It's overexposed and gives few clues as to whether the dress is in cool, bluish daylight or shadow, or under warm, yellowish indoor light. Assume bluish light, and the brain removes blue from the image, leaving white and gold. Assume yellowish light, and it removes yellow, leaving blue and black. Rosa Lafer-Sousa and colleagues showed this directly: placing the dress in scenes with obvious warm or cool lighting swung most viewers to one reading or the other (Lafer-Sousa, Hermann & Conway, 2015). The demo does the same with two backgrounds.

Who sees what

The largest study of the split came from Pascal Wallisch, who surveyed more than 13,000 people online. The strongest predictor was what people assumed about the lighting: those who thought the dress had been photographed in shadow were much more likely to see white and gold. Wallisch also asked whether people were morning types or night owls, and morning people were more likely to see white and gold. His explanation is that people who spend more of their waking hours in daylight, where shadows are bluish, are more practised at discounting blue light, while night owls spend more time under yellowish artificial light and discount that instead. Age and gender made comparatively little difference (Wallisch, 2017).

No single factor explains the split, and it has nothing to do with good or bad eyesight: people with completely normal colour vision end up on both sides. The photo just happens to be almost perfectly ambiguous about the one thing, the colour of the light, that the brain normally guesses without anyone noticing.

Does it hold up here?

The poll under the demo asks Wallisch's two questions: what colours you see in the original photograph, and whether you're a morning person or a night owl. Answer after looking at the photo on Wikipedia, and the breakdown table shows whether this site's visitors repeat his pattern, with more white-and-gold answers among morning people than among night owls. It's a small, self-selected sample rather than a controlled study, so treat it as a fun replication rather than evidence. Still, if the pattern shows up even here, that says something about how robust his finding is.

Some people find their view of the photo flips over time, and many find that a cropped or colour-corrected version, which gives clearer clues about the lighting, can push them to the other side. Others never see anything but their first impression.

A wedding on Colonsay

Cecilia Bleasdale took the photo in a shop in Cheshire while choosing an outfit for her daughter's wedding on the Scottish island of Colonsay. Her daughter and future son-in-law disagreed about its colours, and in February 2015 a friend of the couple posted it online. Within hours it was one of the biggest viral debates the internet had seen, drawing in celebrities, news outlets and vision scientists, and the journal Current Biology published three papers on it within months (Brainard & Hurlbert, 2015; Gegenfurtner, Bloj & Toscani, 2015; Lafer-Sousa, Hermann & Conway, 2015). It relies on the same light-guessing as simultaneous contrast and the checker shadow; it's simply the rare case where the guess came out two ways instead of one.