Baby eye colour predictor

The honest version of a fun question

Almost every pregnancy involves this conversation at some point, usually late at night and usually with a phone in hand. What colour eyes will the baby have? It is a lovely thing to wonder about, and the internet is full of tools that will answer it with confident-looking precision: 87% brown, 11% green, 2% blue.

Those numbers are made up. Not approximate, not simplified — made up. There is no published table of parent eye colour against child eye colour that anyone could derive them from, and the genotype frequencies you would need to build one vary enormously between populations. So this tool does something different. It runs the genetics properly, shows you every square of the working, and gives you a range with the reason for its width attached.

How the classical model works

The two-gene model treats eye colour as the output of two switches. The first, in the OCA2–HERC2 region of chromosome 15, decides brown or not brown, and brown wins whenever it is present. The second, on chromosome 19, decides green or blue among people who did not get brown, and green wins over blue.

So brown eyes need only one brown allele from either parent. Green eyes need no brown allele at all plus at least one green. Blue eyes need the recessive option at both genes — four specific alleles, two from each parent.

That last point explains the thing everyone has heard about. Two brown-eyed parents can absolutely have a blue-eyed baby, and it happens whenever both parents are carrying a hidden blue allele at each gene. The odds under the model are one in sixteen. It is ordinary recessive inheritance, it is not evidence of anything else, and this tool will show you exactly which square it comes from.

Why grandparents narrow the answer

The width of the range comes entirely from what cannot be seen. A brown-eyed person might carry two brown alleles or one, and their face gives nothing away. Every extra bit of family information collapses some of that uncertainty.

The most useful single fact is whether a parent has a blue-eyed parent of their own. A blue-eyed person carries the recessive option at both genes, so they can only have passed on a blue allele at each. If your mother has blue eyes and you have brown, you are carrying one brown and one not-brown allele with certainty. That is why this tool asks, and it is one of the reasons the calculators that do not ask cannot be giving you a real number.

Where the model breaks, and by how much

The two-gene story is a teaching model. It is elegant, it is genuinely useful for explaining recessive inheritance, and the size of the simplification is now measurable.

In 2021 a genome-wide association study of up to 192,986 people identified 124 independent associations across 61 separate regions of the genome, 50 of which had never been reported before. Together those common variants explain about 53% of the variation in eye colour. The authors note in passing that eye colour was, in the past, assumed to be genetically simple. It is not. It involves genes for melanin production, but also genes affecting the structure of the iris itself.

Green and hazel are where the simplification hurts most, and the evidence for that is stark. IrisPlex is a validated forensic tool that reads six variants straight from a DNA sample. Validated in a Slovenian population, it identified blue eyes almost perfectly and brown eyes very well — and at its standard reporting threshold it identified intermediate eye colours with a sensitivity of zero per cent. If reading the actual DNA cannot reliably call green, two parents' eye colours certainly cannot. That is also why hazel and amber are not options in this tool: offering them would imply a precision that does not exist.

One gene does most of the work

There is a reason the brown-or-not half of the model holds up better than the green-or-blue half. A single variant, rs12913832 in HERC2, carries most of the predictive information for blue against brown — in a Polish sample of 1,020 people it alone reached an area under the curve above 0.8. It regulates how much of the pigment gene OCA2 gets switched on in the iris.

So the first switch in the classical model is a reasonable stand-in for something real. The second switch is a placeholder for a great deal of complexity that nobody had characterised when the model was taught.

The colour on the day is not the final answer either

Even a perfect genetic prediction would have a timing problem. Many babies are born with lighter eyes that darken over the first year or two as melanin accumulates in the iris. The colour you see in the first weeks is not necessarily the colour your child will keep, and there is no reliable way to tell from a newborn photograph which way it will go.

What this is, and what it is not

It is entertainment with real genetics behind it, and it is honest about the join. It is not a medical test, it is not a genetic test, and it is emphatically not a paternity test. Eye colour outcomes that look surprising are usually recessive inheritance doing exactly what recessive inheritance does, and this calculator exists partly to show you the square it came from.