DNA Test Accuracy Explained: What 99.9% Actually Means

“DNA tests are 99.9% accurate” is one of the most repeated and most misleading claims in consumer genetics marketing. It is technically true for one specific thing the test does, and meaningless for several other things buyers actually care about. Accuracy depends entirely on which part of the test you are asking about.

Genotyping accuracy: this is the 99.9% number

When a lab reads a SNP position on a genotyping chip, the call (which letter you have at that position) is highly accurate. FDA submissions and peer-reviewed studies of the major consumer arrays have consistently shown per-position accuracy above 99% and typically above 99.9%. This is the number the companies are quoting when they advertise accuracy.

This number tells you that if a test reports you are AA at a given SNP, you almost certainly are AA at that SNP. It does not tell you anything about whether the interpretation built on top of that genotype is accurate.

For the underlying mechanics, see how DNA tests work.

Ethnicity estimate accuracy: probabilistic, panel-dependent, revisable

Ethnicity estimates are not measurements. They are statistical comparisons of your DNA to a reference panel of people with documented ancestry from specific regions. Two consequences follow.

First, the estimate is only as good as the panel. Reference panels for the British Isles and Western Europe are very large and well-characterized. Reference panels for many African, Asian, Indigenous American, and Pacific Islander populations are smaller, sometimes much smaller. Estimates in well-covered regions are correspondingly more precise.

Second, the estimate changes when the panel changes. AncestryDNA, 23andMe, and MyHeritage all periodically update their reference panels, and your percentages can shift meaningfully from one update to the next. This is normal and not a sign that the original estimate was wrong; it reflects the panel improving. Two different companies will also give you different estimates from the same DNA, because they use different panels and algorithms. That is also normal.

What the estimate does well: tell you that your DNA broadly resembles populations from certain regions. What it does not do: assign you to a specific tribe, ethnic group, or nationality. We cover what to do with the numbers in our guide on reading your DNA test results.

Relative matching accuracy: real, but database-bound

When a test says you share 425 centimorgans with someone, that measurement is reliable. Centimorgans, abbreviated cM, are an objective measurement of shared DNA segments and the major companies measure them consistently.

The relationship prediction built on top of that measurement is a range, not a single answer. 425 cM is consistent with second cousins, first cousins twice removed, half-first cousins, and several other relationships. The test will narrow the options; only family history can usually pick the right one.

The other limit on matching is the database. A real cousin who tested at AncestryDNA will not show up in your 23andMe matches, ever, because the two companies do not share data. This is a major reason database size and upload acceptance matter when you are choosing a test. AncestryDNA’s roughly 25 to 30 million customers is much larger than any other consumer service as of early 2026. MyHeritage and FamilyTreeDNA accept free raw-data uploads from other companies, which is the standard workaround.

Health report accuracy: it varies, by a lot

Consumer health reports are the area where accuracy claims need the most scrutiny.

The FDA has authorized 23andMe to report specific carrier-status results and a defined set of health predisposition reports. Those authorizations mean the FDA reviewed the analytical accuracy (does the test correctly identify the variant) and the clinical validity for the specific variants covered. The 23andMe reports do not cover every variant associated with each condition, only the specific ones authorized.

For other consumer health-reporting products (tellmeGen, SelfDecode, various wellness-focused services), regulatory status, analytical accuracy, and clinical validity vary. Marketing claims about variant counts and “99% accuracy” should be read as the company’s claims rather than independent fact.

A critical point about all consumer health reporting: a negative result does not rule out a condition. Most genetic conditions are caused by many possible variants, and consumer tests only check a subset. A clinical genetic test ordered through a doctor and reviewed by a genetic counselor is a different category of product entirely.

If a consumer health report concerns you, do not use the test result alone to make medical decisions. Talk to a board-certified genetic counselor via the NSGC directory or your physician.

The accuracy question to actually ask

Instead of “is this test accurate”, the more useful question is “accurate at what”. The same kit can be highly accurate at genotyping, reasonably accurate at large-region ethnicity, less accurate at fine-grained ethnicity, reliable at measuring cM, only as good as its database for matching, and limited to specific authorized variants for health. All of those things are true at once.

For the bigger picture on picking a test, see our guide to choosing a DNA test.