Red, green and blue became the primary colours on every screen because the human retina carries just three types of cone cells — the pixels aren’t matching the light, they’re matching your eyes

Red, green and blue became the primary colours on every screen because the human retina carries just three types of cone cells — the pixels aren't matching the light, they're matching your eyes

Hold a magnifying glass to a switched-on phone and the illusion falls apart. The smooth photograph on the screen resolves into a tight grid of red, green and blue dots, each one glowing at a different brightness. There is no yellow subpixel, no cyan, no orange.

A sunset on a display contains no orange light at all — only red and green photons hitting the eye close enough together that the brain, unable to tell them apart, invents the orange. The screen isn’t matching the colour of a real sunset. It’s matching the response of three specific cells in the back of the retina, and everything else follows from that.

Human colour vision runs on cone photoreceptors, and there are only three kinds. The short-, medium- and long-wavelength cones, usually labelled S, M and L, peak in sensitivity at roughly 420, 530 and 560 nanometres — very approximately blue, green and yellow-green.

Each cone doesn’t report a colour; it reports a rate, a count of photons weighted by its own sensitivity curve. The brain sees only those three numbers and reconstructs the entire visible world from the ratios between them. Every colour a person has ever seen, from a kingfisher’s back to a bruise, is a triangle of firing rates on those three channels.

Three numbers, not a spectrum

This is why screens can get away with what they do. If human vision sampled the full spectrum — the way a prism spreads it — a display would need to reproduce every wavelength faithfully. Instead, vision throws almost all of that information away at the retina.

Two lights with wildly different physical spectra can produce identical S, M and L responses and look indistinguishable. Colour scientists call these pairs metamers, and they are the loophole every screen exploits. A monochromatic yellow laser at 580 nanometres and a mixture of red and green LEDs can excite the same ratio of L and M cones. To a spectrometer they are different lights. To a human, they are the same yellow.

The formal recognition of this dates to the 1920s. Physicists William David Wright and John Guild ran painstaking experiments in which observers adjusted knobs controlling the intensity of three coloured lamps until the resulting mixture matched a target hue.

Their combined data became the CIE 1931 colour-matching functions, which the International Commission on Illumination adopted as the mathematical foundation of every colour standard since — from television broadcasting to printer inks to the sRGB space your browser is rendering this paragraph in. The functions are, in effect, a numerical portrait of an average human retina.

Why red, green and blue specifically

Given that the eye has three sensors, a screen needs at least three primaries to address them independently. But why these three?

The choice is a compromise between physiology, physics and manufacturing. To reproduce as many colours as possible, the primaries should sit near the edges of the visible spectrum, spreading the addressable triangle — the “gamut” — as wide as the eye can perceive. Red and blue at the two ends and green in the middle roughly maximise that triangle.

Anything closer to the centre would compress the gamut and desaturate every image. Anything further out runs into the limits of what phosphors, LEDs and quantum dots can actually emit efficiently.

Modern wide-gamut standards like Rec. 2020 push the primaries closer to pure spectral colours — a deeper red near 630 nanometres, a greener green near 532 — precisely to enlarge that triangle, but the underlying logic is unchanged. Three sensors, three primaries, one triangle of everything you can see on that screen.

There is nothing sacred about red, green and blue as physical categories. A hypothetical species with four cone types would need four primaries to be fooled properly, and a two-cone species — including most mammals, dogs and cats among them — would only need two.

Some birds and reptiles have four or even five cone types and can perceive ultraviolet. A television built for a pigeon would look, to a human, like it was missing a channel.

The colours that don’t exist

Because the trick relies on stimulating cone ratios rather than reproducing physical light, screens can produce colours that never appear in nature — and fail to reproduce colours that do. Pure spectral violet, near 400 nanometres, excites the S cones strongly and the L cones very slightly. No combination of a display’s red and blue primaries can quite hit that ratio, which is why the deepest violets in a real rainbow always look a little off on a monitor.

Magenta, on the other hand, has no wavelength at all. There is no single frequency of light that produces the sensation of pink or magenta; the brain manufactures it whenever the L and S cones fire together without much M in between, a pattern that the sun’s continuous spectrum essentially never delivers on its own.

Some rare humans carry a fourth functional cone type — a condition called tetrachromacy, arising from an X-chromosome quirk. In principle they should see distinctions the rest of the population cannot, and metamers that fool the standard three-cone eye should fall apart for them. In practice, the extra cone often overlaps heavily with the existing ones, and rigorous demonstrations of expanded colour discrimination have been rare. Their screens, calibrated for three-cone viewers, would be lying to them about most things — the same way a black-and-white television lied to everyone.

Subpixels, dithering and the persistence of the trick

Zoom back out from the retina to the display, and the engineering follows the biology. LCDs use white backlights filtered through red, green and blue subpixels. OLEDs use three separately emitting organic materials. Quantum-dot displays use nanocrystals tuned to precise wavelengths. All of them are trying to hit the same three cone responses with as much control as possible, and they all use the same sRGB or Display P3 mathematics to decide, for any input colour, how bright each of those three lights should glow.

The compression is astonishing. A single 24-bit pixel — eight bits per channel — encodes roughly 16.7 million distinct combinations, which is close to the number of colours the average human eye can actually discriminate. Not because 16.7 million is a special number, but because it lines up with what three cones can meaningfully distinguish. The bit depth of your monitor is a physiological measurement in disguise.

Everything downstream of that — the JPEG algorithm’s aggressive throwing-away of colour detail the eye barely notices, the way the GIF format squeezed images down to a 256-colour palette in the modem era, the entire discipline of video compression — leans on the same fact.

Human vision is far cheaper to satisfy than to describe. Cameras record spectra, screens emit three lights, and somewhere in the middle a great deal of physical information is discarded because the eye was never going to notice it in the first place.

The pixels on the screen, in the end, aren’t a picture of the world. They’re a picture of what three cells in the back of a human eye happen to be capable of counting.

Subscribe to our newsletter!

Our latest tutorials delivered straight to your inbox

Make Tech Easier Editorial Team Avatar

Read next

Inside a modern processor, transistors switch on and off billions of times a second — and they’ve shrunk to about 3 nanometres wide, roughly 15 silicon atoms across, close to the physical limit of how small a switch can be
In 1987, CompuServe engineer Steve Wilhite built GIF around LZW compression and a palette capped at 256 colours for users on 2,400-bit-per-second modems, and the file structure that emerged still governs the looping images moving through browsers and chats today
How did a single mistyped command by one AWS engineer in February 2017 knock a huge slice of the internet offline for nearly four hours?
Wi-Fi doesn’t actually slow down when a café gets crowded — the signal is fine, but every device has to wait its turn to speak on the same frequency, and the polite queuing is what you feel as lag
In a Milwaukee workshop in the 1870s, Christopher Latham Sholes spent years rearranging the keys of his mechanical Type-Writer so that commonly paired letters sat far apart, ensuring the metal typebars would swing from different directions and not jam mid-word
In 1858, engineers laid the first telegraph cable across the Atlantic — it worked for only three weeks before failing, but it cut the message time between London and New York from ten days to a matter of minutes, and set the pattern for every undersea internet cable since
Noise-cancelling headphones don’t block sound, they generate its exact mirror image — a microphone samples the incoming wave, and the driver plays an inverted copy that meets it and cancels to near silence
When you delete a file, the data itself stays on the disk untouched — the operating system only removes the pointer to it, which is why forensic tools can recover files months after they seem to have vanished