Take the hot pink #FF0A73. Show it to a protanope and it arrives as #555F74, a cool slate with a blue cast. Show it to a deuteranope and it arrives as #9A906C, a warm khaki. Those two results sit 33.4 CIEDE2000 apart, which is not a subtle disagreement. It is roughly the distance between a lemon and a leaf.
Both of those people are red-green color blind. They have the diagnosis in common, the awareness-week graphics in common, and the failed Ishihara plate in common. What they do not have in common is the color they are looking at.
I wanted to know how far that goes, so I ran the Brettel dichromat simulation over the whole sRGB cube and measured it. The two halves of red-green color blindness differ from each other by more than a just noticeable difference across 80.7 percent of all colors. The label describes a smaller split than it hides.
What red-green color blindness actually is
Normal color vision runs on three cone types: short, medium and long wavelength, usually written S, M and L. The L cone peaks in the yellow-green around 560 nm and the M cone peaks a little below it around 530 nm. They overlap heavily. Almost everything you can tell apart in the red-to-green direction comes from the small difference between two signals that mostly agree.
The genes for the L and M photopigments sit next to each other on the X chromosome in a head-to-tail array, and they are about 98 percent identical. That arrangement invites unequal crossing over during recombination, which is why this particular deficiency is so common rather than a rarity. Men carry one X, so a single bad array is enough. Roughly 8 percent of men of European descent and about 0.4 percent of women have some form of it.
There are four named varieties, and the distinction that matters is which cone is affected, not which color sounds like it is missing.
- Protanopia. The L cone is absent. Two cones remain, S and M.
- Protanomaly. The L cone is present but its sensitivity is shifted toward the M cone, so the two signals agree even more than usual.
- Deuteranopia. The M cone is absent. Two cones remain, S and L.
- Deuteranomaly. The M cone is present but shifted toward the L cone. This is the most common form of all, and the mildest.
The two -anopia forms are dichromacies, meaning a cone class is genuinely gone, and they are what a simulation can model cleanly. The two -anomaly forms are anomalous trichromacies, which sit on a sliding scale from nearly normal to nearly dichromatic. Everything measured below is the dichromatic case, so treat it as the severe end rather than the average experience.
How the measurement works
The simulation is the standard one from Brettel, Viénot and Mollon. You convert a color into cone response space, then project it onto the surface of colors a dichromat can still tell apart, which is built from the neutral axis and two anchor wavelengths. The projection needs two half-planes rather than one, joined along the neutral axis, because a single plane misses badly at the ends.
I stepped through the sRGB cube at intervals of 3 on each channel, giving 636,056 colors, and simulated each one as a protanope and as a deuteranope. Differences are CIEDE2000, the current standard for how far apart two colors look, where about 2.3 is the threshold at which a careful observer starts to see a difference at all. The headline figures held to within 0.02 across sample densities from 32,768 up to 636,056 colors, so they are not artifacts of how finely I sampled.
One honest caveat before the numbers. A simulation shows what information survives, not what the experience feels like from the inside. Someone born a dichromat has never had the missing signal, so nothing looks wrong to them. The images below are a translation for trichromatic readers, not a window into another person's head.
Finding one: the two types disagree with each other
The obvious comparison is each simulation against normal vision. Protanopia shifts a median 19.51 CIEDE2000 and deuteranopia 20.07, so by that measure they look like the same condition with the same severity.
The more interesting comparison is the two simulations against each other. If red-green color blindness were one condition, those two outputs would nearly agree. They do not. The median disagreement between what a protanope and a deuteranope receive from the same starting color is 5.87 CIEDE2000, and 80.7 percent of the cube exceeds one just noticeable difference. For 58 percent of colors the gap is over 5, which is the range where people reach for different color names.
#FF000020.4
#FF80007.7
#FFFF001.9
#00FF007.6
#00FFFF8.1
#0000FF7.8
#8000FF11.6
#FF00FF17.2
#FF008033.5
#F2A0B510.4
Look at the rose row. A saturated pink reaches a protanope as a blue-grey slate and a deuteranope as a warm sandy khaki. One reads cool, the other reads warm. If you asked both to sort a pile of socks by color they would produce different piles and each would be internally consistent.
The pattern behind the disagreement is lightness, and it is worth stating plainly because it is the single most useful thing to know about the difference. The L cone contributes most of the luminance signal. Lose it, as a protanope does, and reds go dark. Lose the M cone instead and luminance is barely touched.
Finding two: protanopes and deuteranopes disagree about brightness
Pure red sits at L* 53.2. A protanope receives it at 38.9, a drop of 14.4. A deuteranope receives it at 58.5, which is 5.2 brighter. The two types move red in opposite directions on the lightness axis.
- Brake light red #D8261C: 11.1 L* darker to a protanope, 3.9 lighter to a deuteranope
- Stop sign red #B22222: 9.3 L* darker to a protanope, 3.4 lighter to a deuteranope
- Warning red #E5342A: 10.9 L* darker to a protanope, 3.9 lighter to a deuteranope
This is the practical difference between the two diagnoses and it is the one worth designing around. To a protanope a red warning light is a dim warning light, and a red object against a dark background loses the contrast that was supposed to make it jump out. To a deuteranope the same red stays about as bright as it was and the problem is purely one of hue. Across the whole cube the worst-case darkening is 15.3 L* for protanopia and 3.6 for deuteranopia, a factor of more than four.
Finding three: red and green do not collapse into each other
The name promises that red and green become the same color. They do not, and this is the most common misconception about the condition.
Simulate pure red and pure green for a protanope and you get #6A5B0E and #FFEE00, which are 45.7 CIEDE2000 apart. For a deuteranope they are #A28C00 and #F2D12E, 20.0 apart. Both are enormous compared with a threshold of 2.3. A red-green color blind person can separate a red traffic light from a green one without difficulty.
What changed is which channel carries the difference. Hold lightness fixed and recompute, and 96 percent of the surviving protanope difference and 98 percent of the deuteranope difference turns out to be lightness alone. The hue distinction is gone. What is left is that one of them is darker than the other.
That single fact explains most of the confusing behaviour around this condition. Traffic lights work because they encode the message in brightness and position as well as hue. Ishihara plates defeat the same person because they are built specifically to remove that crutch: the dots are randomised in lightness and size so the only thing separating figure from ground is the hue difference on the damaged axis.
Finding four: the loss is directional, and the other axis is perfect
The clean way to see the deficiency is to start from a neutral grey, step outward in every direction of the a*b* plane at fixed lightness, and ask how much of each step survives the simulation.
The blue-yellow axis survives untouched. Not mostly, not adequately: 99 to 100 percent of the difference comes through. The red-green axis keeps 22 to 28 percent under protanopia and 10 to 12 percent under deuteranopia. Everything in between is a blend of the two, falling off smoothly as the direction rotates.
Deuteranopia is the more severe flattening on the damaged axis, keeping about a tenth of it, while protanopia keeps a quarter but pays for it with the lightness distortion from the previous section. Neither is straightforwardly worse. They fail differently, which is exactly the point.
This directional picture is the opponent process made visible. Color vision is not three independent channels but a light-dark channel plus a red-green channel plus a blue-yellow channel. Red-green color blindness deletes one of those three and leaves the other two intact, which is why the result is so orderly.
Which everyday color pairs actually break
Given all that, the pairs that genuinely become indistinguishable are not the ones the name predicts. I took common pairings and measured the separation under each simulation.
Red against brown collapses completely for a protanope, landing at 1.9, below the threshold where any difference is visible. Blue against purple collapses for a deuteranope at 1.4. Neither pair contains a red-green contrast in the everyday sense of those words, and neither appears on the awareness posters.
The blue and purple result is the one people find hardest to believe, and it follows directly from the axis chart. Purple is blue plus a red component. Delete the red-green channel and the only thing separating purple from blue goes with it. Violet #8000FF reaches a deuteranope as #006CFE, a plain blue.
Sorting the whole cube by how far each color moves confirms that the name is pointed at the wrong region. Median shift by hue, for colors above 25 percent chroma:
- Magenta and rose, 300 to 360 degrees: 29.9 to 33.9. The largest movers by a clear margin.
- Red and orange-red, 0 to 30 degrees: 26.0 to 28.3
- Green, 120 to 150 degrees: 25.8 to 26.6
- Yellow and warm orange, 30 to 60 degrees: 5.1. Almost untouched.
- Azure, 210 to 240 degrees: 3.6 to 5.4. Almost untouched.
Magenta and pink move further than red or green do. This mirrors what I found when measuring tritanopia, where the label says blue-yellow and the color that actually dies is yellow while blue comes through nearly intact. Both names were coined from the confusion axis rather than from which colors visibly change, and in both cases that makes the common-sense reading of the name wrong.
What this means if you are designing something
The rules that fall out of the measurements are narrower and more useful than the usual advice to avoid red and green.
- Vary lightness, not just hue. The light-dark channel is completely intact for both types. A red and a green at the same L* are a coin flip. The same two at 25 L* apart are unambiguous for everyone.
- Blue against yellow is the safe pairing. It is the surviving axis and it comes through at 99 to 100 percent. Blue against orange works nearly as well.
- Watch purple. Purple and blue as two categories in the same chart is a worse choice than red and green, and nobody warns about it.
- Do not rely on saturated red for anything urgent. For a protanope it arrives 10 to 14 L* darker than you designed it, so red-on-dark loses contrast exactly where you wanted attention.
- Red and brown is a real trap. It measures below threshold for protanopes, and it turns up constantly in charts, heatmaps and status badges.
Can it be fixed
No, not in the sense people mean. A missing cone class cannot be restored with a filter. The tinted glasses sold for this work by notching out part of the spectrum where the L and M responses overlap, which exaggerates whatever difference remains between two signals that already mostly agree. That can make some contrasts pop, and it can also distort colors that were being seen correctly. I went through what the evidence supports separately in the piece on color blind glasses. Gene therapy has restored trichromacy in adult squirrel monkeys and remains experimental in humans.
If you want to know where you sit, a plate test is the standard screen and you can take one on the color blind test page. Bear in mind that a screen test cannot control your display calibration or your room lighting, so treat a borderline result as a reason to see an optometrist rather than as a diagnosis.
The short version
Red-green color blindness is one name over two conditions that differ from each other by a median 5.87 CIEDE2000, exceeding a just noticeable difference across 80.7 percent of sRGB. Protanopia darkens red by up to 15 L*; deuteranopia leaves brightness alone and flattens hue harder. Red and green do not merge, they keep a large difference of which 96 to 98 percent is pure lightness, which is why traffic lights are readable and Ishihara plates are not. The blue-yellow axis survives at 99 to 100 percent while the red-green axis keeps 10 to 28 percent. The colors that move furthest are magenta and rose, not red or green, and the pairs that actually collapse are red against brown and blue against purple.
All of that is about telling two colors apart while they sit in front of you. Holding one in your head and matching it later is a different and harder problem, and it does not spare anyone their cone complement. That is the gap the Color Memory Game is built on. If you want the trichromatic version of the same question, there is how many colors you can see, and for the other kind of dichromat there is what colors dogs can see, which is close to a natural deuteranopia.