Was the Dress Blue and Black or White and Gold?

Cecilia Bleasdale photographed the dress in a Roman Originals shop at the Cheshire Oaks mall, just north of Chester, in early February 2015. She bought it, for £50, to wear at her daughter Grace’s wedding. When she texted the photo to Grace, her daughter replied, “Oh, the white and gold one?” Bleasdale said it was blue and black. Her partner, Paul Jinks, who had held the dress in the photo, also saw white and gold.

Grace shared the photo privately with friends on Facebook. After the wedding in Scotland, one of them, Caitlin McNeill, posted it on her Tumblr blog, and the next day, 27 February, it went viral. BuzzFeed’s first page on the photo was seen by 26.5 million people between Thursday evening, US east-coast time, and 11am the next morning, the Guardian reported.

The argument was about one photo seen in different ways, not about taste. To some people it showed a white dress with gold lace, and to others a blue dress with black lace. Taken on their own, the pixels of the dress are blue and brown, according to Rosa Lafer-Sousa and colleagues, who studied the photo in 2015. People in their laboratory, looking at a calibrated screen, still split.

Two large surveys disagree on which view was more common, and they asked in different ways. Lafer-Sousa’s team concluded that showing a reduced image biased their respondents toward blue and black. In a follow-up, the share who saw white and gold rose as the image got bigger. Pascal Wallisch asked people online to recall what they saw when they first met the photo. In his survey, white and gold was the majority view.

The rest gave other answers. In Lafer-Sousa and colleagues’ survey, 11% said blue and brown; in Wallisch’s, the most common other answer was blue and gold.

Download this finding: which version most people saw

This section as one image. Its bars show Lafer-Sousa and colleagues’ survey of 1,401 people and the first run of Wallisch’s survey, of 8,084 people. In Lafer-Sousa and colleagues’ survey, 57% saw blue and black and 30% white and gold. In Wallisch’s first run, 59% saw white and gold and 27% blue and black. The image is 1,200 by 630 pixels, and anyone may download it and post it.

Download the card (PNG, 1,200 × 630)

The visual system judges color by judging the light, and this photo leaves the light unclear. The light that reaches the eye from a surface depends on both the surface and the light shining on it. The visual system discounts the light to keep an object’s color stable, an ability called color constancy. To do that, it has to judge what the light is. Manuel Melgosa and colleagues note that this photo lacks the information to tell, and people fill the gap in different ways.

The colors of the dress make the judgment harder. Karl Gegenfurtner, Marina Bloj and Matteo Toscani found that they lie close to the daylight locus, the range of colors that daylight takes over a day, from yellow to blue. That makes it hard to tell a change in the light from a change in the fabric. The bright blue tones could come from a white dress in dim bluish light, or from a blue dress in bright neutral light.

Lafer-Sousa’s team proposed that people who assume cool, bluish light, such as a blue sky, discount the blue and see white and gold. People who assume warm light, such as an incandescent bulb, discount the yellow and see blue and black. To test the idea, the team placed the dress in a scene lit to look warm and in another lit to look cool. Most people then saw the dress the way the lighting predicted.

The blue in the photo matters. Alissa Winkler and colleagues found that people more often see a surface as white or gray when its color varies toward blue than toward yellow. They suggest people tend to put bluish tints down to the light.

When Winkler and colleagues inverted the photo’s colors, so the blue stripes turned yellow, 94% of 87 students called the stripes yellow or gold and none called them white. Gegenfurtner’s team rotated the colors of the dress away from blue and yellow, and none of their 15 observers still called it white.

White and gold is what the photo shows if the dress is a white one in bluish shade. Pascal Wallisch found that people who thought the dress was in shadow were more likely to see white and gold, strongly so in both runs of his online survey. Shadows are bluish, he notes, so assuming one would lead the visual system to discount the blue. White and gold was also the majority view in both runs of his survey, at 59% and 56%.

Andrey Chetverikov and Ivan Ivanchei found a second cue. In their online study, white-and-gold viewers were more likely to think a light was behind the dress, and to read the bright background as a window. Lit from behind, the dress might be underexposed, and discounting that would make its colors look lighter than they are.

Age and sex make some difference, but the studies disagree on how much. In Lafer-Sousa’s survey, women and older people were more likely to see white and gold, and the authors call age and sex the two most significant predictors in their study. Wallisch found the difference between men and women small and statistically unreliable. In his survey the white-and-gold share rose into people’s 40s and fell sharply after about 65. Both teams raise the yellowing of the eye’s lens with age as a possible factor, without testing it.

Three accounts of the physical dress agree on its colors: the retailer’s, the buyer’s and a measurement of a bought dress. Roman Originals, the retailer that sold it, tweeted on 27 February 2015: “We can confirm #TheDress is blue and black!” Bleasdale, who bought it, told the Guardian: “I’ve never seen the white and gold. Ever.”

Melgosa and colleagues bought one of the dresses and measured how much light its blue and black parts reflect at each wavelength. Under standard daylight, the readings correspond to the colors usually called blue and black. The authors add that, strictly, objects have no color of their own, since color depends on the light falling on them, among other things.

Blue-and-black viewers seem to assume warm or direct light. Lafer-Sousa’s team suggests they take the light to be warm, like an incandescent bulb, and discount the yellow. Chetverikov and Ivanchei found that 47% of blue-and-black viewers saw a light in front of the dress, against 19% of white-and-gold viewers. Lit from the front, as by a camera flash, the dress might look overexposed, and discounting that would make its colors look darker.

Most of the studies agree that people’s assumptions about the light drive the split. Wallisch writes that the field seems to converge on this explanation. Christoph Witzel and colleagues replicated it in two online surveys in 2017. Aston and Hurlbert found that the color people judged the light to be explained variation in the colors they matched to the dress. To make a match, people adjusted a color on screen until it looked the same as part of the dress.

Some findings do not fit neatly. Chetverikov and Ivanchei found that people sitting in warmer room light, from incandescent, halogen or light-emitting diode lamps, more often saw white and gold. They note that this goes against the discounting explanation. Aston and Hurlbert found that people’s skill at color constancy, measured with a standard task, was unrelated to how they saw the dress. Wallisch cautions that no one has made another image as ambiguous, which suggests the effect is not fully understood.

Researchers also disagree on whether people see three distinct versions or a range. Lafer-Sousa’s team found three peaks in the colors people matched, which they read as three stable views: white and gold, blue and black, and blue and brown. In a study of 15 observers, Gegenfurtner’s team found a continuous range instead. Their two groups differed mainly in how light or dark they saw the dress.

Whether sleep habits play a part is not settled. Lafer-Sousa’s team and Wallisch both suggested that early risers, who are likely to see more bluish daylight, would come to expect bluish light. In both runs of Wallisch’s survey, respondents who called themselves morning types more often saw white and gold than those who called themselves night owls.

Aston and Hurlbert, testing 32 people in the laboratory, found moderate links with a questionnaire on morning and evening types. The links were significant only at the 10% level. Lafer-Sousa’s own survey found only a slight trend that was not significant.

The time of day people looked is less certain still. In Chetverikov and Ivanchei’s first study, among 318 people in Moscow, the dress was more often seen as blue and black in the evening and at night. The rest of their sample showed no such effect. Their second study did not find the pattern.

Brain studies of the dress are few and small. Lara Schlaffke and colleagues scanned 28 people, 14 in each group, and found extra activity, mainly in frontal and parietal areas, in those who saw white and gold. Wallisch calls that study relatively low powered, meaning it had less chance of detecting a real effect. Talia Retter and colleagues recorded brain waves from 27 people who said which version they saw. From activity recorded over the back of the brain, they could tell blue-and-black from white-and-gold viewers with over 80% accuracy.

How often people switch between versions differs between studies. In Lafer-Sousa’s laboratory, 3 of 28 people retested a few days later reported a different version. Among their online respondents who had seen the photo before, 45% said their view had switched at some point. In Wallisch’s survey, most respondents said theirs had never switched.

Researchers also differ on how far the dress is like images that flip between two views. Lafer-Sousa’s team likened it to multistable images such as the Necker cube. Chetverikov and Ivanchei draw the opposite contrast: unlike the Necker cube, they write, changes in how the dress looks are rare and limited to part of the population.

Knowing the true color pulls people toward it. In Wallisch’s survey, the few who knew the dress was blue and black when they first saw the photo were more likely to see it that way. Most of those whose view later changed switched from white and gold to blue and black.

Context can change what people see. Besides the size of the image, Lafer-Sousa’s team found that blurring the photo made blue and black more likely. At demonstrations for a vision science meeting in 2015, Aston and colleagues lit the real dress with mixed yellow and blue light. Most observers saw it as blue and black against a yellow background, and as white and gold against a black one.

Melgosa and colleagues draw a general lesson: color is a human perception, and the visual system’s answer often rests on assumptions of unknown and variable origin. Wallisch suggests that when the light is unclear, people’s color experiences may differ more than vision scientists had assumed.

Lighting also affects memory for faces. In a 2022 laboratory study by Denise Lim and colleagues, people learned faces and were then tested on them. Changing the brightness of the light in between made them worse at recognizing the faces, and the authors point to eyewitness testimony. The Innocence Project, a legal organization that works to free wrongly convicted people, counts 375 wrongful convictions in the US overturned by DNA evidence from 1989 to 2020. Mistaken eyewitness identification played a part in 69% of them.

Blue and black. The dress Cecilia Bleasdale photographed is a blue dress with black lace, as the retailer that sold it confirmed and as measurements of a bought dress show. White and gold is a reading of the photo rather than of the fabric, and in the larger of the two surveys it was the more common reading.

The photo divides people because it carries too little information about the light on the dress, and its colors run along the range that daylight takes. In the studies, clear lighting cues flipped most viewers, and people who thought the dress was in shadow leaned toward white and gold. Why individuals assume different light, whether through sleep habits, age or something else, is not settled.

Download the conclusion

The verdict above as one image. Its bars show how many people switched in Lafer-Sousa and colleagues’ 2015 study when the dress was placed in a scene with clear lighting. Of those who first saw white and gold, 64% saw blue and black under warm light. Of those who first saw blue and black, 80% saw white and gold under cool light. The image is 1,200 by 630 pixels, the size social networks use for link previews, and anyone may download it and post it.

Download the card (PNG, 1,200 × 630)