Screens make almost every color from just red, green and blue light. So is white light really made of red, green and blue? And why does yellow light plus blue light make white, when yellow paint plus blue paint makes green? Let's find out.
Shine sunlight through a prism and it spreads out into a rainbow. That's because light comes in different wavelengths. Think of them as waves of different sizes, measured in nanometers (nm). A nanometer is tiny: a human hair is about 80,000 nm thick. Violet has the shortest waves your eyes can see (about 380 nm). Red has the longest (about 700 nm).
The rainbow has no gaps. It's a smooth slide through thousands of slightly different wavelengths, and sunlight contains all of them. The chart below is a spectrum. Read it like this: short waves on the left, long waves on the right, and the height shows how much of each wavelength the light has. Try the buttons, or draw your own light and see what color it makes.
✏️ Drag on the chart to draw how much of each wavelength your light has.
The striped strip on the far left is ultraviolet (UV): waves too short for your eyes to see.
This box shows the color, not how bright it is. Your eyes adjust to brightness.
Sunlight has every wavelength, roughly evenly. All of them together look white.
The back of your eye is covered in light sensors called cones. Here's the surprise: there are only three kinds, named after the waves they like best: S for short, M for medium and L for long.
The chart shows how strongly each kind responds to each wavelength. The higher its curve, the harder it fires. Slide a single wavelength along the rainbow and watch all three cones react.
Each bar is as tall as the dot on its curve
At 450 nm (blue): L hardly fires at all, M hardly fires at all, S fires a lot. Your brain reads that pattern of three as blue.
Notice that the curves are wide and overlap. Most wavelengths tickle two cones at once, just by different amounts. That mix is how your brain tells colors apart. Notice too that S's curve is tall and narrow, while L's and M's are lower but wider. That evens things out, as you're about to see.
Real light, like sunlight, has lots of wavelengths at once. A cone can't keep track of them all separately. Instead it does something simple: it adds them all up into one number. Watch how.
The thin line on each bar marks 100: what sunlight scores. Tap a bar to watch that cone add up.
The white line is sweeping across the rainbow. At each wavelength, the M cone takes how much light there is and multiplies it by how much it likes that wavelength (the dashed curve). If the cone only half-likes a wavelength, it counts half of that light. So lots of light where the curve is high makes a tall column, and little light, or a low curve, makes a short one. Each new column drops onto the pile in the M bar.
So each cone reports just one number: its total. Your brain never gets the whole spectrum, only those three totals. Every color you have ever seen is just a different set of three cone totals.
It's like describing a whole song with only three volume meters: one for low notes, one for middle notes and one for high notes. Lots of different songs could give the same three readings. That's the loophole we're about to use.
Your brain only gets three numbers. So two completely different lights that give the same three numbers must look exactly the same. Scientists call pairs like this metamers.
Let's fool your eye. On the left is a mystery light. On the right you have just three narrow lights, red, green and blue, like the tiny dots on a screen. Each slider turns one light up or down. Can you mix them so they look identical? You'll know you've done it when your three bars reach the dashed lines.
Each bar is built from the light that fills it, just like in part 2, and the number is its total. The dashed lines on the right are the mystery light's totals: that's your target. The thin line on every bar marks 100, what sunlight scores. Both spectrum charts use the same scale, so you can compare how much light each one has.
Your S bar is too high. Blue light mostly tickles S, so turn blue down. Stuck? Press Show me and watch the sliders move.
So is white light made of red, green and blue?
No. Sunlight is made of every wavelength. Red + green + blue just happens to give your three kinds of cone the same three numbers, so to a human it looks the same. That's the only reason TVs, phones and computer screens can get away with three colors of light. They are built for human eyes. A bird would need a screen with four colors of light. For a dog, two would do. “Red, green and blue” isn't a fact about light. It's a fact about you.
When you shine lights on top of each other, their light adds up. Each light drops its own piece into your cone bars, so the bars get taller and the patch gets brighter. This is called additive mixing, and it's how screens work. Look at a screen through a magnifying glass and you'll see tiny red, green and blue dots.
Tap or drag on the dark wall to move the ✛ and measure the light there. The bars show which light filled them.
Drag the little rings to move the lights. Tap or drag anywhere else to put the ✛ there and measure the light.
White, but only to human eyes. This spot has just three narrow colors of light, nothing like the full rainbow in sunlight. But all three cone bars come out equal, just like sunlight makes them. And since red + green looked yellow, you could also say: yellow light + blue light = white!
Paint, crayons, leaves and bananas don't make any light of their own. White light, with every wavelength in it, lands on them. They absorb (eat) some wavelengths and bounce the rest into your eye. The color you see is whatever is left over:
white light − what the paint eats = what you see
Paint can only ever take light away, so this is called subtractive mixing. Pick a paint and watch.
The dashed line is all the white light that arrives. Striped = eaten by the paint. Rainbow = bounced back to your eye.
Yellow paint eats violet and blue, and bounces back green, yellow, orange and red. That leftover light is what your eye sees.
Look closely at yellow. Yellow paint doesn't just bounce yellow light. It bounces green, yellow, orange and red. That mixture gives your cones a very similar pattern to pure yellow light (L and M high, S low), so it looks yellow. Now click blue: blue paint bounces violet, blue and a good amount of green. Remember that. It's about to matter.
In part 4, the yellow patch (red + green light) plus blue light made white. Lights add. But you've probably mixed yellow and blue paint and got green. That's because mixing paint doesn't add light. Each paint eats its own part of the white light, and the mix eats what both paints eat:
white light − what yellow eats − what blue eats = what you see
Lots of people learn a simple idea: “light is red, green and blue, so yellow paint bounces red and green, and blue paint bounces blue.” If that were true, what would yellow + blue make? Try the cartoon paints first, then switch to real paints.
Each bite here can be a little smaller than in that paint's own chart above. In the mix, each paint is only part of the paint, and light that one paint has eaten can't be eaten again.
Black?! That's what the simple idea predicts. If yellow paint only bounced red and green light, and blue paint only bounced blue light, nothing would be left. Yellow eats the blue and blue eats the rest. But you've mixed yellow and blue paint, and you know it makes green. So the simple idea must be wrong somewhere. Switch to Real paints.
Mixing paints is like stacking colored filters. Each paint takes its own bite out of the rainbow, and only light that every paint lets through makes it back to your eye. The simple idea chops the rainbow into three chunky blocks with nothing in common. Real paints have smooth, wide curves that overlap, and the overlap, the light both paints bounce, is where the green comes from. Try the other pairs too.
Here's the strangest part. A color's name tells you your three cone totals. It doesn't tell you which wavelengths are actually there. So two paints can look identical but bounce completely different light, just like the mystery light and your three lights in part 3.
Blue A bounces a wide, gentle hill of light. Blue B bounces a tall, narrow spike near 485 nm, a little violet and a small bump of red. So how can they possibly look the same?
Remember part 2? At each wavelength, a cone takes how much light there is (the charts in step 1) and multiplies it by how much it likes that wavelength (the dashed curve). That makes a column. Then it piles up all its columns into one total. The cone can't tell which columns made the pile, only how big it is. (A sheet of white paper would score about 90.)
Blue A spreads its violet and blue in a wide hill. Blue B gets about half of its S total from the blue around 450 nm and half from its tall spike. Different columns, same pile.
Blue A tickles M mostly with its long tail into the greens. Blue B has less of that tail, but its spike sits on the edge of the M curve and makes up the difference.
Blue A's green tail also tickles L a little. Blue B makes up the same amount with the edge of its spike plus its small bump of red light.
Same three totals, so the same color. Look inside the bars: Blue A and Blue B fill them with different light, but each pile ends up the same height. Your brain only ever gets the three totals, so the two blues look identical. The difference is still there in the light. Your eye just has no way to see it.
The two blues look exactly the same. Guess: will they make the same color with yellow?
Every light and paint is a spectrum sampled every 2 nm from 360–720 nm. Human color uses the CIE 1931 color-matching functions (Wyman, Sloan & Shirley 2013 analytic fit), converted to cone responses with the Hunt-Pointer-Estevez matrix and normalized so an equal-energy spectrum gives L = M = S = 100. The stacked bars split each cone response into 10 nm slices of the spectrum, so the slices sum exactly to the reading. Displayed colors adapt that equal-energy white to sRGB white (von Kries) and clip out-of-gamut channels.
Dog (429/555 nm) and bird (370/445/508/565 nm) cones use the Govardovskii visual-pigment template without lens or oil-droplet filtering, so they are approximate. Paint mixing uses single-constant Kubelka–Munk theory: absorption-to-scattering ratios add by concentration. The “eaten by” bands split each wavelength's absorbed light between the paints in proportion to their share of the mix's K/S. Paint reflectance curves are illustrative shapes, not measured pigments. The two look-alike blues are a solved metameric pair under equal-energy light.