How Accurate Are Ethnicity Estimates? An Honest Assessment

“How accurate is the ethnicity estimate” is the single most common question we get about consumer DNA tests, and the honest answer has two parts. Continental-level estimates are very reliable. Sub-regional estimates are educated guesses, and how educated they are depends entirely on whether the company has enough reference samples from that part of the world. The big companies are not equally accurate, and accuracy is not equally distributed across regions even within one company.

What “accurate” means in this context

An ethnicity estimate is a statistical inference, not a measurement, so the right question is not “is it right” but “what fraction of windows of your genome are assigned to the correct region.” The companies publish accuracy data in their white papers, and the numbers are useful even if you have to read them carefully.

AncestryDNA’s published accuracy data shows continental-level assignments correct in well over 95% of test cases, which matches independent analysis. Sub-regional accuracy is much lower and varies by region. The same is true at 23andMe and MyHeritage, with the specifics differing by which regions each company has invested in.

Why continental-level estimates are reliable

Two genomes from anywhere in Europe differ from two genomes from anywhere in East Asia in many small ways across many SNPs. The signal is large, and even a small reference panel can detect it. This is why every major company gets the continental breakdown right for almost every customer, and why a “100% European” or “99% Sub-Saharan African” continental result is one you can trust at face value.

The exception is people whose ancestry crosses continental boundaries within the last few generations. Latino and Hispanic customers, for example, often see breakdowns that mix European, Indigenous American, and African ancestries. The continental categories are correct, but the proportions depend heavily on reference panel size for the Indigenous American component, which has historically been smaller.

Where sub-regional estimates fall apart

Sub-regional estimates depend on reference panel size, diversity, and correctness within a region. Three regional categories illustrate the problem.

British Isles sub-regions, including England, Scotland, Ireland, and Wales, share enormous amounts of recent ancestry. Telling Scottish from Northern Irish ancestry inside a window of your genome is hard, and the major companies disagree noticeably about a typical customer’s breakdown. AncestryDNA’s UK panel is the largest in the industry, which is why it became the de-facto standard test for UK-focused genealogy.

African sub-regions remain undersampled industry-wide. West African, East African, and Central African categories are still much broader than European categories, and “Nigerian” or “Senegalese” sub-regions cover populations far more genetically diverse than the labels suggest. Researchers and customers from African ancestry have been clear that the industry has a long way to go here.

Indigenous American reference panels are also small, partly for historical research-ethics reasons and partly because the populations involved are small. Many customers see “Indigenous Americas” as a single broad category where European customers would see five or six sub-regions.

What this means for how you use your estimate

Treat the continental breakdown as solid. Treat sub-regional percentages as “the algorithm’s best guess given the panel,” and read the confidence intervals if the company shows them. The expanded view at AncestryDNA, the range view at 23andMe, and the v2.0 detail at MyHeritage are all worth clicking into. A “37% Scottish” estimate with a range of 22-49% is telling you something very different from a “37% Scottish” estimate with a range of 33-41%.

When two siblings get different sub-regional breakdowns, neither is wrong. Siblings inherit different sets of DNA from the same parents, and small sub-regional differences are exactly the noise floor at which the algorithm is working. We cover the underlying mechanism in understanding ethnicity estimates, and the periodic refreshes in why ethnicity estimates change.

Which company is most accurate

There is no single winner. AncestryDNA leads on British Isles and broader European sub-regions because its panel for those regions is the largest. 23andMe has historically led on European Jewish ancestry and has invested in expanding African and East Asian reference panels. MyHeritage’s v2.0 update improved European granularity and remains useful for customers who want to compare two companies’ estimates from one sample.

For genealogy work, the matching side of a result is almost always more informative than the ethnicity estimate. See how DNA matching works and the ancestry pillar guide for the broader picture, or pick a test from our comparisons hub.