The 1920s experiment behind every screen colour

Every hex code on your screen is a point on a map, and that map was drawn from the colour matches of just 17 people. Ten of them turned knobs for David Wright at Imperial College, London. The other seven did the same for John Guild at the National Physical Laboratory. The two labs worked separately, on different machines, and they still agreed. Their result became the international standard in 1931.

Why does a hex code mean the same colour everywhere?

Because a hex code is a recipe, and the units it's written in were fixed in 1931. When you type #0FA3B1 into Figma, every browser reads it as an sRGB colour, which means so much red, so much green and so much blue. Those three ingredients, and the white they add up to, are pinned to exact points on one map. And that map was built from what 17 people saw when they matched colours by hand.

So your teal means the same thing in Sydney as it does in Stockholm, all because 17 volunteers matched colours by hand in the 1920s.

The student who built the machine

W. D. Wright, known as David, was a research student in his early twenties at Imperial College's Technical Optics Department. In 1927–28, on a Medical Research Council grant, he designed a colour-matching instrument and made much of it himself in the college workshop. He even worked in the room that Abney and Watson had designed for their own colour research.

His three lamps were pure rainbow lights, at about 650 nm for red, 530 nm for green and 460 nm for blue, and he picked them for practical reasons. Now, the observer looked at a field about 2° across. Wright chose that size on Guild's suggestion, because it keeps the image on the very centre of the eye. Most diagrams draw the field as a circle. Wright's was a square, though, split into two rectangles. One half showed a test colour and the other showed a mix of the three lamps.

What did the volunteers actually do?

They turned three drums, using both hands, until the two halves matched. Each drum pulled a string, and the string slid a grey wedge across the red, green or blue beam. Wright wanted the drums to turn smoothly "to ensure the peace of mind of the observer." I like him for that line.

It took two people to run each session. One looked, and the other set the test colours and wrote down the readings. A single match took a minute or more, and observers were told to look away and check again before they trusted it, so a full set took several hours. Wright admitted that recruiting volunteers was "a rather thankless task."

Ten people matched the whole spectrum for him, and dozens more did spot checks. Most of the differences between those people, he found, came from the yellow pigment at the centre of the eye. But while his paper names his two assistants, it doesn't name his ten observers. Guild's seven are unknown too.

Maxwell got there first, in a box. Back in 1860 James Clerk Maxwell ran colour matches of his own, with an observer calling for more or less red, green or blue. His wife was one of those observers, and he valued her readings especially. He did it all in a black viewing box nearly eight feet long, and his neighbours thought he was mad for "staring into a coffin" for hours. Like Wright, he blamed the eye's yellow spot for the differences between people.

Where "minus red" comes from

For about a third of the rainbow, from violet-blue through cyan to green, no mix of the three lamps could make a match. The rainbow colour was simply purer than anything the lamps could make together. So the observers did something clever. They shone red onto the rainbow side instead, which dulled the target until the mix could reach it.

On paper, red added to the target gets written down as minus red. Nothing about the light was negative, only the bookkeeping. In the form the standard took in 1931, red goes negative from 436 to 546 nm. Green and blue also dip below zero in places, by tiny amounts, so that third of the rainbow is where the big dose of red was needed, not the only place anything went negative.

The dip comes from the eye, not the equipment. A later study in 1955 used different lamps and found the same thing. It's also the reason no screen shows every colour you can see: sRGB covers only about a third of the full map, which Why your brand colour looks different on your phone picks up. That's why a vivid cyan you've seen in the world can come out duller on screen.

How did two labs with different machines get the same answer?

Because both labs were measuring the same thing, which is the human eye. Over in Teddington, at the National Physical Laboratory, John Guild had built his own colour-matching machine back in 1922. It used filtered light rather than pure rainbow lamps, and in 1926–27 he measured seven observers on it.

Some explainers say Wright's lab and Guild's were one team, but they weren't. When Wright published in 1929, he didn't even know which three primaries the NPL used, and he said so himself, in print. Guild then asked him, in public, to restate his results in the NPL's terms. Wright did that in 1930.

Once both sets were in the same units, they agreed. Guild summed it up as different groups of observers, using "entirely dissimilar apparatus." And that agreement is the whole reason the result could become a standard.

A week in Cambridge, September 1931

From 13 to 19 September 1931, the CIE (the international body for light and lighting) met in Cambridge. It passed more than seventy resolutions on lighting of all kinds, and only a handful of them fixed the colour standard.

At first the standard was written in three real lamps: 700, 546.1 and 435.8 nm. The last two are mercury lines, which any lab can reproduce easily. But that version still had the minus red in it, and the Americans balked. In Smith and Guild's words, "American opinion is unfavourable to the use of negative numbers."

So the committee re-expressed everything in three imaginary lights, called X, Y and Z, that no lamp can make. The colours on the map didn't change, only the reference they're measured from, and now no point on the map ever sits below zero. Several of the choices were made simply to save work for people doing sums on desk calculators, to seven decimal places. Y was made identical, value for value, to a 1924 brightness curve that came from a separate US study of 52 people.

The meeting itself comes to us second-hand. Wright told the story later, and Fairman, Brill and Hemmendinger relayed his account in 1997: the committee met at Trinity College at 9:30 a.m. on 18 September, with 21 delegates from six countries. Smith, Guild and Priest were said to have stayed up most of the night computing the tables. Wright later said hardly anyone in the room understood one of the resolutions. That same 1997 paper concludes that none of the 1931 design principles would likely be chosen if the system were built fresh today. We use it anyway.

How do 17 people's matches reach your screen?

Through a chain of standards, and every link in it points back to 1931. The link to screens started early, too. Wright himself spent 1929–30 at Westinghouse in Pittsburgh, working on colour television. Today the chain runs like this:

I rebuilt the conversion numbers in the CSS colour standard from four 1931 points, sRGB's red, green, blue and white, and they match exactly, fraction for fraction. Newer colour spaces, Oklab included, are calculated from XYZ too.

Because Y is that 1924 brightness curve, Y is what decides how bright your colours count as. Pure blue, #0000FF, comes out at about 7% as bright as white, but pure green comes out at about 72%. Two colours that look equally strong to you can be a long way apart in brightness, and brightness is what a contrast check measures.

A side note: your eye isn't red, green and blue. The three cone types in your eye peak at about 570, 543 and 442 nm, which is yellow-green, green and violet-blue. The so-called red cone actually peaks in yellow-green. X, Y and Z aren't the cones either. Colour-matching curves and cone sensitivities carry the same information in different units, because matches are decided at the cones.

Where the 1931 standard falls short

Wright and Guild only ever measured ratios between their three lamps. The full curves were rebuilt from those ratios using that 1924 brightness curve, which, in the words of a UCL vision lab, "seriously underestimates" the eye's sensitivity below 460 nm. The error came along for the ride, which means the 1931 standard is a ruler that's a hair short at the blue end. And values below 400 nm and above 700 nm were never measured at all. They're extrapolations.

For decades that hardly mattered. Older screens spread their light across broad bands, where a slightly short ruler averages out. But newer OLED and laser screens squeeze their light into narrow peaks, and sometimes those peaks sit right where the ruler is short.

One more thing: the 1931 observer was built for a thumbnail. It's meant for colour patches about 1° to 4° across, which is roughly a thumbnail at arm's length. For bigger patches the CIE added a second observer in 1964, built from 67 people. Hex codes, sRGB and ICC colour profiles all still use the 1931 observer, though.

Why can two calibrated screens disagree about white?

Because the 1931 observer is an average, and screens with narrow-spectrum light show up where it's off. Sony ran into this with its first OLED master monitor in 2011. Its OLED and CRT monitors didn't match "even though they were calibrated to have the same xy chromaticity values." So Sony now ships its OLEDs deliberately set off the standard white. It also found that no single replacement observer fixed the problem, because individual eyes differ. And a small 2020 study found that phone OLED screens set to the 1931 numbers tend to look green-tinted.

Then in April 2026, Apple went further. Its Studio Display XDR is calibrated by default to "Apple CMF 2026", and Apple cites "limitations in the standard colorimetric system derived from the CIE 1931" curves. The 1931 version is still there in its reference presets. According to Apple's support page, daylight white sits at x 0.3144, y 0.3302 under the new curves, against 0.3127, 0.3290 under 1931. What Apple hasn't published is what its curves are based on.

Apple says it's working on the problem through the CIE's own committee, JTC 22, whose remit names "imperfections in the CIE 1931 standard observer". But nobody has replaced 1931. It's still the international standard, and the CIE's newer observers, from 2006 and 2015, are still technical reports under verification.

If the white on your phone and the white on your monitor don't quite agree, even after calibrating, this is likely why.

17 people, seven decimal places

The standard counts "a total of 17 observers", ten at Imperial College and seven in Teddington. Their matches were averaged, reshaped by hand arithmetic to fit a brightness curve from 1924, and frozen to seven decimal places. In other words, the standard observer isn't anybody. Nobody's arm is exactly a metre long, and nobody's eye is the standard observer either.

It has held for 95 years because it's very nearly right. And every colour you've ever picked, you picked in their units.

What should you do differently in your next design?

Use hex codes as exact recipes, but expect screens to be close rather than identical.

1. Hand over colours as hex or sRGB values, never as "that teal". The numbers carry the same 1931 meaning in every browser, so they're the one part of a handover nobody can misread.
2. Don't judge brightness by how strong a colour looks. Pure blue is about 7% as bright as white, pure green about 72%. Check contrast with numbers, not your eye.
3. Expect vivid rainbow colours to fall outside your screen. Some colours you can see in the world can't be shown in sRGB at all. Pick the nearest screen colour on purpose, rather than chasing the one in your head.
4. Sign off colour on the device people will actually use. Two calibrated screens can still disagree slightly about white, especially an OLED phone sitting next to an LCD monitor. Small differences between them aren't your mistake.
5. If your team shares a reference monitor, find out which calibration it uses. A display set to the newer curves and one set to 1931 will show the same white slightly differently.

Frequently asked questions

Who was the "standard observer"?

Nobody. The CIE 1931 standard observer is an average, not a person. It was built from the colour matches of 17 people, ten measured by W. D. Wright at Imperial College and seven by John Guild at the National Physical Laboratory, in two separate studies in the late 1920s. Their results were averaged, adjusted to fit an older brightness curve, and adopted as the international standard in 1931.

Why isn't brown or grey on the chromaticity diagram?

Because the horseshoe diagram leaves out brightness. It only maps hue and how pure a colour is, so brown sits on the same point as the orange it's a dim version of, and grey sits on the same point as white. Mid grey, #808080, lands exactly on white's point. So the diagram shows every hue and purity a person can see, but not every colour. The missing piece is how light or dark it is.

Does a standard from 1931 still affect my design files?

Yes. A hex code is an sRGB colour, and sRGB's red, green, blue and white are defined as points on the CIE 1931 map. Its creators even described sRGB values as combinations of 1931 values. Adobe RGB and every ICC colour profile are anchored to the same 1931 observer too. Some new displays are starting to calibrate to newer colour-matching data, but 1931 is still the international default.

For the curious

The curves the 17 observers produced are called colour-matching functions, and their average is the CIE 1931 2° standard colorimetric observer (ISO/CIE 11664-1:2019). The first form used real primaries at 700, 546.1 and 435.8 nm; X, Y and Z are imaginary primaries, and the functions are written x̄, ȳ and z̄. Think of a balance scale: a weight on the other pan counts as negative, and that's all "minus red" is. ȳ equals the 1924 photopic luminosity function V(λ) value for value, which is why sRGB's luminance row is 0.2126, 0.7152, 0.0722, the same weights WCAG uses. A point on the horseshoe is a chromaticity: #0FA3B1 sits at x 0.2192, y 0.3042. Two screens that match for one person but not another show observer metamerism.

My figures (red negative from 435.8 to 546.1 nm, 36.9% of 400–700 nm; the exact CSS matrix) were computed from the CIE's open data. Guild's own 1932 paper, the CIE's 1931 proceedings and Wright's 1981 account weren't available to me, so the Cambridge scene stays attributed.

Sources (27)

• W. D. Wright, "A trichromatic colorimeter with spectral primaries" (opens in a new tab), Transactions of the Optical Society 29(5), 225–242, 1927–28, with discussion
• W. D. Wright, "A re-determination of the trichromatic coefficients of the spectral colours", Transactions of the Optical Society 30(4), 141–164, 1928–29, with discussion and reply (archive.org)
• W. D. Wright, "A re-determination of the mixture curves of the spectrum", Transactions of the Optical Society 31(4), 201–218, 1929–30, with discussion (archive.org)
• J. Guild, "A trichromatic colorimeter suitable for standardisation work", Transactions of the Optical Society 27(2), 1925–26 (archive.org)
• T. Smith and J. Guild, "The C.I.E. colorimetric standards and their use", Transactions of the Optical Society 33(3), 73–134, 1931–32 (archive.org)
• H. S. Fairman, M. H. Brill and H. Hemmendinger, "How the CIE 1931 color-matching functions were derived from Wright–Guild data" (opens in a new tab), Color Research & Application 22(1), 11–23, 1997
• A. D. Broadbent, "A critical review of the development of the CIE1931 RGB color-matching functions" (opens in a new tab), Color Research & Application 29(4), 2004 (web version)
• K. S. Gibson and E. P. T. Tyndall, "Visibility of radiant energy" (opens in a new tab), Scientific Papers of the Bureau of Standards 19, 1923
• Nature 128, 1083 (opens in a new tab), 1931 (the Cambridge session)
• J. C. Maxwell, "On the theory of compound colours" (opens in a new tab), Philosophical Transactions 150, 1860
• L. Campbell and W. Garnett, The Life of James Clerk Maxwell (opens in a new tab), 1882
• R. W. G. Hunt, "W. D. Wright obituary" (opens in a new tab), The Colour Group (GB)
• Optica, "William David Wright" (opens in a new tab)
• ISO/CIE 11664-1:2019, CIE standard colorimetric observers (opens in a new tab) (preview)
• CVRL, UCL, notes on the CIE 1931 2° colour-matching functions (opens in a new tab), colour-matching functions (opens in a new tab) and cone sensitivities and colour matching (opens in a new tab)
• A. Stockman, "Cone fundamentals and CIE standards" (opens in a new tab), Current Opinion in Behavioral Sciences 30, 2019
• W3C, CSS Color Module Level 4 (opens in a new tab), Candidate Recommendation Draft, 13 September 2026
• M. Stokes, M. Anderson, S. Chandrasekar and R. Motta, "A Standard Default Color Space for the Internet: sRGB" (opens in a new tab), v1.10, 1996
• ITU-R, Recommendation BT.709-6 (opens in a new tab), 2015
• ICC, ICC.1:2022 (opens in a new tab)
• Adobe, Adobe RGB (1998) Color Image Encoding (opens in a new tab), 2005
• CIE, JTC 22, "Optimized Colour Matching Functions for Display Colour Consistency" (opens in a new tab)
• CIE, open dataset: CIE 1931 2° colour-matching functions (opens in a new tab), CC BY-SA 4.0
• Sony, "Color matching between BVM-HX310 and BVM-X300", v1.00, 18 December 2020
• Y. Wei, X. Hu and M. R. Luo, "Observer metamerism to display white point between LCD and OLED displays" (opens in a new tab), IS&T Color and Imaging Conference 28, 2020
• Apple, Studio Display XDR Technology Overview (opens in a new tab), April 2026
• Apple Support, article 105101 (opens in a new tab), 11 March 2026