The rarest eye color, and why no iris can make green

The internet has settled on green as the rarest eye color, and the internet is wrong. The largest count anyone has ever made of human eye color came out in 2025: iris color as recorded on the driving licences of 235,423,085 people across 31 American states. Green came fourth.

Grey is the rarest, by a factor of thirteen over green. That is a satisfying enough answer on its own, but it is not the interesting part. The interesting part is why the list looks like this, and the reason is not genetics. It is optics. So I built an iris and measured it.

Everything below comes out of a physical model of the iris assembled from published tissue-optics values, run through the CIE observer and converted to color the same way the rest of the pieces on this site do it. The headline result surprised me: sweep the only pigment the human eye actually contains across its entire range and you get a single line from blue through grey to brown, sixteen distinguishable colors long, that never once enters the green part of color space.

Nobody has blue eyes

Start with the fact that makes the rest of it make sense. There is no blue pigment in any human eye. There is no green one either, and no grey, and no hazel.

When Wielgus and Sarna ran chemical degradation and electron spin resonance on human irides across the full range of colors, what they found was eumelanin, the same brown-black pigment that colors hair and skin, plus pheomelanin at a level they described as a few percent. That is the entire inventory. Imesch and colleagues put the anatomy the same way after reviewing the morphology: the differences between normal eye colors come from variable amounts of melanin granules held in a constant number of melanocytes in the superficial stroma. Everyone has the same cells in the same places. What differs is how much pigment those cells are carrying.

So the blue is not pigment. It is the same thing that makes the sky blue: short wavelengths scattering more than long ones off structures much smaller than a wavelength of light. In the iris those structures are collagen fibrils. Jacques names the length scales directly in his review of tissue optics, a 70 nm periodic density along the fibrils and gaps between them under 100 nm, which is why collagen-rich tissue carries far more Rayleigh scattering than other soft tissue does.

A blue eye, then, is a faintly cloudy layer over a black backing. The backing is the posterior pigment epithelium, which is densely pigmented in absolutely everybody regardless of eye color, and which is doing the job that the dark of space does for the sky: swallowing everything that is not scattered back out.

How I built it

The model is deliberately plain, because every part of it had to come from a published number rather than from me.

• Scattering. The reduced scattering coefficient of the stroma follows Jacques’ two-term form, a Rayleigh part going as the fourth power of wavelength and a Mie part that is nearly flat. I used his tabulated means for fibrous tissue: 29.2 per cm at 500 nm, with 48.9 percent of it Rayleigh. No ocular tissue appears in that table, so collagen-rich fibrous tissue is the closest honest proxy.
• Absorption. Melanosome absorption follows the same review: 519 per cm at 500 nm, falling as roughly the cube of wavelength, scaled by the volume fraction of melanosomes in the layer. That volume fraction is the one knob I turn.
• Geometry. A 150 micron scattering layer on a perfectly absorbing backing, solved with Kubelka-Munk. Total iris thickness measured by swept-source OCT in 233 healthy adults runs 365 to 372 microns; the anterior stroma is the optically live part of that.

Two things this model does not do, stated up front. It does not reproduce absolute reflectance to better than a factor of something, so I make no claims about how bright any particular iris is. And it treats the iris as flat and uniform, which real irides emphatically are not. What it does give, reliably, is the shape of the color response: which directions in color space the pigment can move you, and how far. Every conclusion below is about direction and distance, and every one of them survives being re-run across the full published spread of the input parameters. I checked all 48 combinations.

The ladder

Here is the whole of human eye color, generated by turning one number. The percentage on the left is the melanosome volume fraction in the anterior stroma.

Blue, then slate, then a dead neutral grey at five percent, then warm, then brown. That is the entire human range and it is one parameter wide. Nobody had to add anything to get from one end to the other. I find it slightly unnerving that a family of colors we treat as categories, the way we treat blood groups, is actually a dial.

A detail worth pausing on: the top rung, the zero-melanin limit, comes out at #707F96. CSS has a name for very nearly that color already, and it is slategray, #708090. Lightness 52.80 against 52.80. The model, given no instruction about what an eye should look like, lands on the word English speakers reach for when they describe pale eyes.

It is mostly a brightness axis

The distance from the blue end to the brown end is 32.67 CIEDE2000 units. Now hold the chromaticity of the blue end fixed and change only its lightness to match the brown end.

• Lightness change alone: 27.48 units of the 32.67.
• Chromaticity change alone: 17.67 units.
• L* falls by 32.81 across the range; the a*b* coordinates move only 22.03.

Eighty-four percent of the perceived difference between a blue eye and a brown eye is that one is darker than the other. This keeps happening. It is the same result I got measuring red-green color blindness, where red and green stayed apart for dichromats almost entirely on lightness, and it is the same reason the dress splits people. Hue does much less work in real scenes than our vocabulary implies.

The pigment does almost nothing until it does everything

The ladder is not evenly spaced, and the unevenness is severe. Doubling the melanin load moves the color by wildly different amounts depending on where you start.

• 0.05% to 0.1%: the iris moves 0.22 units. Invisible.
• 0.5% to 1%: 2.08 units. Right at the threshold.
• 2% to 4%: 6.67 units.
• 5% to 10%: 11.57 units. Unmistakable.

The same proportional change in pigment is fifty times more visible at the brown end than at the blue end. Which predicts something you can check without any equipment: blue eyes should all look much the same as each other, and brown eyes should range enormously, from honey through chestnut to almost black. They do.

It also quietly resolves a disagreement in the literature. Koblova and colleagues, fitting an inverse Monte Carlo model to in-vivo iris images, put mean melanin at 26.2 mg/mL for blue irides against 30.3 for brown, a difference of sixteen percent. Wielgus and Sarna, measuring by electron spin resonance, found brown irides about forty percent higher than everything else. Neither figure sounds anything like enough to turn a blue eye brown, and both are whole-iris averages dominated by the epithelium at the back, which is fully pigmented in everyone. The color lives in a thin anterior layer that holds a small share of the total. A modest difference in the whole organ can hide a very large one in the part that we actually look at.

Turning off the blue

If the blue really is structural, removing the Rayleigh term should destroy it and leave everything else standing. So I set the Rayleigh fraction to zero, keeping the total amount of scattering the same, and re-ran the ladder.

Take the Rayleigh away and the blue eye loses 71 percent of its chroma and becomes a slightly cool grey. That is the whole of the blue, confirmed by removal.

The brown eye does the opposite. Without Rayleigh it gets four times more saturated, chroma 5.74 to 22.63. The scattering that makes blue eyes blue is still running in brown eyes, fighting the pigment and washing it toward neutral. Every brown eye is a blue eye with the blue drowned out, and the drowning is not total.

Your eyes and the sky are the same blue

The zero-melanin iris and the zenith sky I computed in the sky piece are made by the same mechanism, so it is fair to ask how close they land.

Five degrees of hue apart. The raw difference between them is 22.85 units, but raise the iris to the sky’s lightness and it collapses to 8.25, and what is left is purity rather than hue. The sky is a pure scatterer with nothing behind it. The iris has a flat Mie component diluting the Rayleigh, which is why it never gets as saturated. Same color, thinned.

A blue eye is barely blue

Here is the spectrum doing the work. The palest iris in the model reflects 29.60 percent at 450 nm and 16.38 percent at 650 nm, a ratio of 1.81. That is all the blue there is: less than twice as much short-wavelength light as long. A brown iris runs the same asymmetry backwards, 3.38 percent against 5.34, a ratio of 0.63.

To see how weak that is, compare it against an iris with a genuine blue pigment in it, one with a real absorption band. A pigment reflecting a 45 nm band centred on 460 nm gives #006DC6, chroma 57.1. The structural blue manages 13.95. A true blue pigment would be four times more saturated and sit 19.1 units away. Nobody has ever seen a blue eye in that sense, and if you did it would look like a contact lens, because that is exactly how colored contact lenses are made.

Green is not on the map

Now the part that made me go back and check the code three times.

Melanin comes in two forms and I modelled both, eumelanin and the yellower pheomelanin, with the spectral shapes from Sarna and Swartz’s monomer extinction coefficients. I swept both across four decades of loading, 48,400 combinations. Then I measured the maximum chroma the model could reach in each region of hue.

The iris can put 32.5 units of chroma into brown and 14.0 into blue. Into green it manages 1.32, which is less than a single just noticeable difference. Against a neutral grey of the same lightness, the greenest thing the pigment can build is indistinguishable from grey.

And it is not a quirk of my parameter choices. Across all 48 combinations of scattering strength, Rayleigh fraction and layer thickness spanning the published ranges:

• Maximum chroma toward blue: 8.14 to 19.36.
• Maximum chroma toward brown: 12.20 to 40.16.
• Maximum chroma toward green: 0.00 to 1.73.
• Combinations where green stayed under one JND: 48 out of 48.

There is a reason this is so absolute, and once you see it you cannot unsee it. Melanin absorbs monotonically: more in the blue, steadily less as you go red. Rayleigh scattering is monotonic the same way. A reflectance spectrum built by dividing one monotonic falling curve by another is itself monotonic. Monotonic spectra make blue at one end and yellow at the other and pass through flat grey in between. To make green you need a bump in the middle, which means absorbing at both ends of the spectrum, and nothing in the iris does that. Green is on the other side of the blue-yellow axis from everywhere this machinery can reach.

Layering does not rescue it, and neither does mixing

Two obvious escape routes, both tested and both closed.

Put the yellow on top instead of mixing it in. The anterior border layer really is a distinct sheet over the stroma, so a yellow filter over a blue scatterer is anatomically plausible, and a filter multiplies rather than averages. This does help. It lifts the best reachable green from chroma 1.32 to 4.22, and the greenest respectable color in the whole search is #5D4D2A at hue 85.0 with chroma 22.9. But hue 85 is not green. The very least green color English has a common word for is CSS olive, and olive sits at hue 102.7. Darkolivegreen is at 121.5, green itself at 136.0. The iris runs out 17.7 degrees short of the nearest green name, and 51 degrees short of green.

Mix brown and blue patches spatially. Real irides are mottled, and at reading distance a mottled iris averages. So I averaged a blue spectrum and a brown spectrum in every proportion from nought to a hundred. The result walks from #707F96 through #4D4E53 to #413A33, which is to say straight down the middle of the same old axis. The greenest point on that entire path had a chroma of zero in the green band. Averaging two colors that sit on a line gives you a point on the line.

So what colour are green eyes

I want to be careful here, because green-eyed people exist and I am not about to tell twenty-one million Americans that they are imagining it.

What the measurement says is narrower than that. It says the optics of the iris, as currently published, cannot produce a color that a colorimeter would place in the green region. Green eyes are real as a category; they are not green as a coordinate. The colors the model reaches at the green end of its range, #62573B and #5D4D2A, are dark olives and khakis, and when you see one set in a face, ringed by white sclera, next to warm skin, with a black pupil beside it, your visual system has plenty of reason to call it green.

That leaves one unresolved possibility, which I will flag rather than paper over. Popular sources routinely attribute green eyes to a yellow pigment called lipochrome sitting over a blue stroma. The chemical work on actual irides, both the ESR and the HPLC studies, finds eumelanin and pheomelanin and does not report a third pigment. If lipochrome is real and abundant, it is not in those results, and my model has no way to include it. If it is not real, then green eyes are exactly the layered olive the model produces and the word is doing the rest.

Why an eye looks more colored than it measures

There is one more effect large enough that ignoring it would make everything above misleading, and it cuts in the direction people actually experience.

An iris is never seen against a neutral background. It is seen ringed by sclera and surrounded by skin, both of which are warm. Adapt the observer to that surround with a standard CAT02 transform rather than to a neutral white, and the same physical iris reports a different color.

The surround multiplies apparent chroma by 1.63 for the pale iris and by 2.12 for the middling one, and the effect gets stronger, not weaker, against darker and warmer skin: 1.76 and 2.36. Blue eyes genuinely do look about twice as blue as a colorimeter says they are, and the effect is largest on the eyes that measure closest to grey. This is color constancy machinery running in the direction it was never designed for, and it is why photographing someone’s eyes and then looking at the crop on a white screen is always a disappointment.

Note which way it pushes: toward blue, every time. Warm surround, cool-shifted percept. It makes pale eyes bluer. It does not make anything greener.

Back to the rarest

Now the census figures have a mechanism behind them. Take the ladder, throw away the rungs too dark to read as anything but black, and measure how much of the remaining journey each color name owns.

• Blue, hue 240 to 290 and visibly colored: 39.7 percent of the axis.
• Neutral grey, chroma under one JND: 21.3 percent.
• Brown and amber: 39.1 percent.
• Green: 0.0 percent.

The optics hands out two large territories and one wide neutral corridor between them. The population does not fill them evenly, which is the genetics part: a single common variant near the HERC2 gene, rs12913832, predicts blue versus brown well enough on its own to explain about two thirds of the ordinal variation in European samples, and it acts by suppressing the pigment gene rather than by tuning it smoothly. That pushes people toward the ends. The middle of the ladder, the genuinely neutral part where an iris has enough melanin to kill the blue and not enough to look brown, is a narrow target in pigment terms and almost nobody lands on it. Hence 0.7 percent grey.

Green, meanwhile, gets 9 percent of a census despite occupying 0.0 percent of the color space. That is not a contradiction; it is a measurement of the word rather than of the eye. Grey is rare because the pigment window is narrow. Green is common because the category is generous.

How many eye colors are there, really

Walk the visible part of the ladder in steps of 2.3 CIEDE2000, the threshold at which two colors stop being the same color, and you get 16 steps along an arc 37.7 units long.

Sixteen. Against a working English vocabulary of about six, which means every eye-color word is covering roughly three distinguishable shades and two of the six words are arguing over territory that contains no distinct color at all. That is the same compression I keep finding everywhere: 67 distinguishable greys against a handful of names, hundreds of thousands of distinguishable colors against a vocabulary of maybe thirty. Naming is lossy, and eye color is one of the places we notice the loss, because people care about the answer and the form only has six boxes.

What I would want to do next

The weak point in all of this is the scattering input. There is no ocular tissue in the standard table, so I proxied with fibrous tissue, and Jacques himself notes in print that even the melanin absorption anchor is contested, with three values in circulation that differ by a factor of 1.3 and exponents from 3.0 to 3.48. I ran the sensitivity sweep because of that, and the green result held everywhere, but a real measurement of iris stroma scattering would settle it properly. The in-vivo hyperspectral work exists; the numbers are in figures behind paywalls rather than in downloadable tables.

The other thing worth doing is checking the corneal problem. Anyone measuring iris reflectance from a photograph is measuring the cornea’s four percent Fresnel reflection as well, which is why the careful studies use confocal rejection. Photographed eye color and iris reflectance are not the same quantity, and a fair amount of the disagreement about what color anyone’s eyes are probably lives in that gap.

The short version

• Grey is the rarest eye color at 0.7 percent, not green at 9, across 235 million people.
• There is no blue, green or grey pigment in any human eye. There is melanin, and there is scattering.
• The full range of human eye color is one pigment at different doses, running #707F96 to #34291D, and 84 percent of the difference between the ends is brightness.
• Remove Rayleigh scattering and a blue eye loses 71 percent of its chroma. Brown eyes get four times more saturated, because the scattering was diluting them all along.
• The iris reaches chroma 32.5 toward brown, 14.0 toward blue, and 1.32 toward green, which is under one JND. It falls 17.7 degrees of hue short of the least green color English has a word for.
• Sixteen distinguishable colors, six words for them.

If reading that has made you curious about how well you can actually judge a color rather than name one, that is the whole premise of the color memory game: you see a color, it disappears, and you rebuild it from memory while CIEDE2000 scores the gap. The Name That Color variant is the version that puts the vocabulary problem right in front of you. And if this piece has you wondering about your own eyes rather than everyone else’s, the eye test and the color blindness test are both here and both free.