An ancient ancestry test does not test anything ancient. It takes the raw file a testing company already produced for you and compares it with genomes recovered from archaeological skeletons, tens of thousands of which have now been published. The result says which ancient populations your genome resembles, and in what proportions, era by era. That sentence is the whole product; the rest of this guide explains what each word in it means, which methods produce the numbers, and what a result can and cannot claim. The ancient DNA test page covers Ancestrify's own version; this post is about the category.
What "ancient ancestry" means genetically#
Your family tree doubles every generation. Thirty generations back, around the year 1200, you have more ancestral slots than there were people alive in your region, so your ancestors are a large slice of the population of wherever your lines were living. Go back three thousand years and the slice is most of a continent. "Ancient ancestry" is therefore never a line of descent from a person or a tribe. It is a statement about population resemblance: which sampled ancient groups, in which era, carry the allele frequencies that best explain the allele frequencies in your file.
That reframing matters because it sets the honest limits. A result cannot say "you descend from the Yamnaya". It can say "a model with 45% Yamnaya-related ancestry fits your genome and a model without it is rejected", or "your coordinates sit closest to Bronze Age Central Europeans", or "you share stretches of DNA with this Iron Age individual at a rate consistent with shared population ancestry". Those are three different sentences, produced by three different methods.
The three methods every commercial test uses#
Formal modelling with qpAdm. qpAdm, from the ADMIXTOOLS family, asks whether your genome can be written as a mixture of chosen source populations, checked against a set of outgroup populations that do not participate in the mixture. It returns proportions with standard errors and z-scores, and a p-value: a model that fails is rejected outright. This is the only one of the three that can say no. It is also the most demanding: the sources must be chosen well, the outgroups must actually separate them, and a model that passes is only as good as the alternatives it was tested against. Ancestrify's qpAdm analysis publishes the p-value, every source's weight, standard error and z-score and the full right set, and a person composes and checks every model; the is qpAdm worth it post covers the difference from a calculator in full.
Global25 distances and admixture. Global25 is a 25-dimensional coordinate system built from a PCA of ancient and modern genomes. Your coordinate row is compared with the rows of reference populations by plain Euclidean distance, and admixture proportions are found by fitting your row as a weighted average of source rows. It is fast, visual and reproducible, and it always returns an answer: the fit distance says how good the answer is, but nothing is rejected. The official row comes only from Davidski's independent Global25 service; Ancestrify never computes coordinates itself, and its Global25 analysis runs on the official row across six eras with every source panel published, used and rejected sources alike.
Shared-segment matching against ancient individuals. Instead of populations, this method compares your file with each ancient genome one at a time and records the stretches where you carry the same alleles. Reported as identity by state, it produces a ranked list of individual people, each with a burial site and a date. It is the most vivid of the three and the easiest to over-read: a long shared stretch with a Bronze Age Briton is evidence about a shared ancestral population, not proof that this person is your ancestor. Ancient Matches paints each match's segments on your chromosomes and states the evidence per row; the IBS explainer says exactly what such a match is.
One genome, read three ways#
Take a customer with four grandparents from northern Italy and run the same file through all three methods. The qpAdm report offers a model of three sources for the Bronze Age era, say an Anatolian Neolithic farmer group, a Western hunter-gatherer group and a Yamnaya-related steppe group, against a right set of a dozen outgroups. It returns something like 62% farmer, 8% hunter-gatherer and 30% steppe, each with a standard error of a few percent, and a p-value of 0.31. That p-value means the model is not rejected; a competing model that swapped the steppe source for an Iranian-related one came back at 0.002 and was rejected, and the report shows both. The Global25 report, on the official row, lists the closest ancient populations in the same era by distance, say a Bell Beaker group from northern Italy at 0.028 and a Bronze Age group from the Alps at 0.031, and fits the row as a mixture of the panel with a stated fit distance. The Ancient Matches report lists the two hundred ancient individuals who share the most DNA with the file, the longest stretches painted on the chromosomes, with an Iron Age burial from the Po valley near the top. Three reports, one genome, three different sentences, and the honest reading uses all three: the model says which mixture is consistent with the data and which is not, the distances say where in the reference space the row sits, and the matches say which individuals carry the same alleles over long stretches. None of them says that the Iron Age burial is a relative.
The eras a result is split into#
Most tests split the reference panel by period, because the same region was home to different populations at different times. In Europe and West Asia the layers are:
- Hunter-gatherers, the foraging populations present before farming; the hunter-gatherer ancestry test post covers this era.
- Neolithic farmers, who spread from Anatolia into Europe from around 7000 BCE and mixed with the foragers they met.
- Steppe pastoralists of the Pontic-Caspian grassland, whose expansion after 3000 BCE reshaped most of Europe and parts of Asia.
- Bronze and Iron Age populations, where the three earlier layers have already mixed and regional differences emerge.
- Classical antiquity, the Greek, Roman, Persian and contemporary worlds, when long-distance movement adds new components.
- Medieval populations, the migration period and the kingdoms that follow, closest to present-day structure.
A good result shows the same genome at several of these depths. A hunter-gatherer proportion in the earliest era is expected to disappear into a Bronze Age source in a later era, because the Bronze Age source already contains it; a result that shows 15% hunter-gatherer in every era is describing the panel, not you.
How to read a result honestly#
Read the fit before the proportions. For qpAdm that is the p-value and the standard errors; for a coordinate fit it is the distance; for a match list it is the segment length and count. A proportion of 12% with a standard error of 9% is a proportion of "somewhere between zero and a fifth", and a report that hides the error is hiding that. Then read what was offered and rejected: a percentage without the list of populations the model could have used is half a result, because a two-source model will always find two sources, and the interesting fact is which third source made the fit worse. Then check that the era matches the question; a medieval-era model cannot tell you about the Neolithic, and a Neolithic-era model cannot tell you about the migration period, because the populations that would carry that signal were not in the panel. Check the reference panel's coverage for your region as well: the published record is dense for Europe and West Asia and thin for much of Africa, the Americas and Oceania, and a sparse panel produces confident-looking results that lean on whatever neighbours exist. Finally, treat every population name as a label for a reference sample, not for a people who could have known your family. If the report names a population as your ancestor, it is selling a story.
Two habits help. Compare eras against each other rather than reading one in isolation: the same genome should tell a consistent story as the panel moves forward in time, with earlier components folding into later ones. And compare methods against each other: a source that qpAdm accepts, that the coordinate fit also uses and that the match list is full of is a robust finding; a source that appears in only one of the three is a question, not an answer.
Which providers do what#
Read on each provider's own site on 2026-09-15; where something was not stated there, it says so. The best ancient DNA test guide has the longer comparison and the compare pages the cell-by-cell facts.
| Provider | Method as stated | Statistics shown | Price as listed on 2026-09-15 |
|---|---|---|---|
| Ancestrify | qpAdm against AADR v66, hand-checked; Global25 on the official row; Ancient Matches by segment | p-value, SE, z-score, right set; fit distance and full panels; per-match segments | qpAdm from 29.99 EUR, Global25 29.99 EUR, Ancient Matches 29.99 EUR, one-time |
| MyTrueAncestry | Comparison with 185 ancient civilizations and ancient samples; statistical method not stated | Not stated | Free Commoner level; paid levels 37 to 844 USD |
| Illustrative DNA | DeepAncestry: PCA-based coordinates, distances to 585 ancient populations across six periods, two and three-way fits; AdmixLab: DIY qpAdm | Genetic fit per model; AdmixLab tool output | Not stated on the pages read |
| Genomelink Ancient Ancestry | Admixture fractions across seven ancient components, algorithm credited to a team led by Carlos Bustamante | Proportions | Free report offered as a limited-time promotion; report prices not stated |
| MyHeritage Ancient Origins | Comparison with ancient populations from the Neolithic to the late Middle Ages, built with Illustrative DNA; 32 notable individuals | A similarity score per population | Premium feature, Complete or Omni subscription |
| DNAGENICS | Admixture, similarity, IBS segments, PCA and UMAP, haplogroups against 9,000+ ancient samples | Not stated | Packs listed at 90, 140 and 160 EUR, shown at 59, 91 and 104 EUR |
Two structural notes. Only the qpAdm-based products (Ancestrify's report, Illustrative DNA's AdmixLab, Genoplot's paid tiers) return a test that can fail; the rest return a ranking or a proportion that always exists. And only Ancestrify runs on the official Global25 row rather than a coordinate space of the provider's own.
A buyer's checklist#
- Does the result include a statistic that could have rejected it? If not, it is a ranking.
- Are the source populations listed in full, including the ones the model did not use?
- Is the reference panel named and versioned, so the result can be re-run?
- Is the era of each result stated, and does it match your question?
- Are coordinates official or the provider's own simulation?
- Is the price one-time, a subscription or a level system, and what does each unlock?
- Where is the file hosted, and can you delete it yourself?
- Does the site say, somewhere, that no method can identify an individual ancestor?
Frequently asked questions#
Can a DNA test tell me my ancient ancestors?#
It can tell you which ancient populations your genome resembles and, with qpAdm, whether a particular mixture of them is statistically consistent with your file. It cannot name an individual ancestor: three thousand years back, your ancestors are a large fraction of the people then living in your regions of origin, and no method can pick one out of that crowd.
What is the best ancient ancestry test?#
For a tested model with the statistics published and a person checking it, Ancestrify's qpAdm analysis from 29.99 EUR. For a free first look, MyTrueAncestry's Commoner level or Genomelink's promotional report. For a sandbox to run your own models, Illustrative DNA's AdmixLab. The best ancient DNA test guide sorts them by question.
Is an ancient ancestry test the same as an ethnicity estimate?#
No. An ethnicity estimate from 23andMe, AncestryDNA or MyHeritage compares you with living reference populations over the last few centuries. An ancient ancestry test compares the same file with genomes from excavated individuals, thousands of years old, and reports by era.
How accurate is an ancient ancestry test?#
As accurate as its statistics say. A qpAdm model states its p-value and standard errors, and a rejected model is the method working. A coordinate fit is accurate to its panel and its fit distance. A segment match is accurate to its length and marker density. None of them is accurate about a named ancestor, because none of them measures that.
Terms used here are defined in the glossary.
