Methodology & sources
Where the numbers come from
U.S. figures on this site come from real datasets. Worldwide figures are estimates, published with a range and a confidence level. This page explains how we set them, including what the data cannot tell us. Sources last checked November 2026.
Our current global estimates
Worldwide point estimates are our judgment, not counts. Ranges describe the plausible spread, not statistical confidence intervals. The figures do not add up to exactly 100%.
| Color | Global estimate | Range | U.S. data | Confidence |
|---|---|---|---|---|
| Brown | ~75% | 70–80% | 45–53% | high |
| Blue | ~9% | 8–10% | 24–27% | moderate |
| Hazel | ~5% | 4–8% | 10–18% | moderate |
| Amber | ~3% | 1–5% | Not separately tracked | low |
| Green | ~2% | 1–4% | 9% | moderate |
| Gray | ~1% | 0.5–3% | 0.7–1% | low |
How each number is calculated
- 01
Aggregate
We collect figures from the sources listed below and trace each one to where it started. Direct counts carry more weight than figures copied between websites.
- 02
Reconcile categories
Where sources classify differently (hazel folded into brown, gray counted as blue), we prefer a wider range to a falsely clean average.
- 03
Publish a range + point
For the U.S., the range runs between the two datasets and the point is the license figure, since that sample is far larger. Worldwide, the point is our judgment within the range. It is used for comparisons and the '1 in X' figure (100 ÷ point).
- 04
Grade confidence
High means every source agrees on the size of the figure. Moderate means sources agree on the rough size, or direct counts exist for part of it. Low means the color is rarely measured directly.
Comparison ratios in the rarity tool are point estimates divided by each other and rounded to one decimal. For example, brown 75% ÷ green 2% = 37.5×.
The source list
Hashem, Menicacci & Granet (2025), PLOS ONE
Visit source ↗Eye color on 235,423,085 driver's licenses from 31 U.S. states, using 2016 records. The largest eye color dataset we know of.
Used for: U.S. point estimates (brown/black 53%, blue 23.7%, hazel 10.3%, green 9%, gray 0.7%) and state figures.
2014 Harris Poll for the American Academy of Ophthalmology
Visit source ↗Survey of more than 2,000 Americans, weighted to the U.S. population.
Used for: The second U.S. dataset: brown 45%, blue 27%, hazel 18%, green 9%, other 1%. Also the explanation of why blue eyes look blue.
Katsara & Nothnagel (2019), Forensic Science International: Genetics
Visit source ↗Literature review of eye color by country, mostly in Europe. We use the table as reproduced by Healthline.
Used for: Country figures. It counts hazel as brown and gray as blue, and groups green with yellow as 'intermediate'.
WorldAtlas and All About Vision
Visit source ↗Popular summaries of global eye color percentages. They give little or no primary sourcing.
Used for: The commonly quoted worldwide figures, which we treat as a starting point and test against direct counts.
Stelzer (1979) and Dabkowski et al. (2022)
Visit source ↗Direct counts of heterochromia: 25,346 people examined in Vienna, and 11,111 West Point cadet portraits.
Used for: Heterochromia prevalence, about 0.26% for all forms and 0.063% for complete heterochromia.
Simcoe et al. (2021), Science Advances
Visit source ↗Genome-wide study of up to 192,986 people of European ancestry and 1,636 of Asian ancestry.
Used for: The polygenic picture: 124 signals in 61 genomic regions, 50 of them new.
Eiberg et al. (2008) and Sturm & Larsson (2009)
Visit source ↗Peer-reviewed work on the HERC2 and OCA2 region and on iris color genetics.
Used for: The origin of blue eyes and how the main genes work.
MedlinePlus Genetics (U.S. National Library of Medicine)
Visit source ↗Government reference on eye color inheritance.
Used for: Inheritance mechanics and gene summaries.
Ludwig et al. (2016) and Bito et al. (1997)
Visit source ↗A newborn eye color study of 192 babies, and the Louisville Twin Study.
Used for: How eye color starts out at birth and how it changes later.
National Organization for Albinism and Hypopigmentation (NOAH)
Visit source ↗Advocacy and reference body for albinism.
Used for: Albinism prevalence and the red or violet eye appearance.
Known limitations
These matter as much as the numbers themselves.
Limitation 01
There is no global census of eye color
Worldwide percentages trace back to popular summaries with little sourcing, not to a measured sample. Asia and Africa hold most of the world's people and have the least data, though brown is close to universal there.
Limitation 02
Categories are fuzzy
Hazel and amber, amber and light brown, and gray and blue are classified differently between surveys, and even by the same person over time. This is the largest source of uncertainty.
Limitation 03
People describe their own eyes
Both big U.S. datasets rely on self-report. Nobody checked the eyes. That is why hazel is 10.3% in one and 18% in the other.
Limitation 04
The U.S. license data has gaps
Only 31 states took part, the records are from 2016, and Arizona and California hold 98.6% of the 'other' group, which is probably missing data. Unlicensed people and under-16s are not included.
Limitation 05
Genetic studies lean European
The largest eye color study included up to 192,986 people of European ancestry and only 1,636 of Asian ancestry, so its findings fit some groups better than others.
Limitation 06
Changes over time
Populations mix and move. We treat estimates as a snapshot and revisit them when new large-scale data appears.
How to cite & correct us
Please cite figures as estimates from Eye Color Rarity with the range and confidence label intact, and link to this page. Example: “An estimated 2% of the world population has green eyes (range 1–4%, moderate confidence), according to Eye Color Rarity.”
Found a primary source we should weigh, or an error? Tell us through the contact page. If a better source changes a figure, we update the page and the date above.