What Is a Reference Population in an Ancestry DNA Test
A reference population is a comparison group of present-day or ancient people whose DNA is used to estimate your ancestry percentages, not proof of a specific direct ancestor.
A reference population is a group of present-day or ancient people whose DNA is used as a comparison set when a test estimates your ancestry. It is a statistical reference group, not proof that one named ancestor belongs to you. On a HealthOSX DNA Kit report, the ancestry panel uses 157 reference populations to estimate percentages across 15 regions, and it reports your closest matches among 298 present-day groups and 133 ancient groups.
What is a reference population in an ancestry DNA test?
The phrase sounds like it should mean the group you descend from. It does not. In autosomal ancestry testing, a reference population is a set of DNA samples from people chosen to represent a geographic or historical group. The company works out how common each DNA variant is in that group, then compares your genotypes against those frequencies.
The basic workflow:
- A lab reads hundreds of thousands of markers, often around 650,000 SNPs, across your genome.
- Software looks at short stretches of your DNA and asks which reference group those stretches resemble most.
- Those stretch-level calls get added up into percentages.
So a result of 23 percent Germanic Europe does not mean 23 percent of your ancestors were German. It means 23 percent of the DNA segments the algorithm could assign looked closest to that reference group.
A reference population is a comparison group, not a family tree. Percentages are model output, not proof of a specific ancestor.
Where the percentages actually come from
Ancestry percentages are not a direct readout. They are an estimate produced by a model.
The model compares your SNPs to allele frequencies in each reference population. Because human populations differ mainly in frequency, not in the presence or absence of variants, the signal is statistical. Your DNA is divided into windows or chunks, and each chunk is assigned to the population that fits best. Some algorithms also look at haplotypes, which are patterns of SNPs inherited together, to make the assignment more stable.
Then the chunks are summed. That is why the output is a percentage. It is the proportion of assigned segments, not the proportion of your family tree. Large chunks are often assigned correctly at the continental level. Smaller chunks, especially for neighbouring populations, are more uncertain.
A panel with 15 regions and 157 reference populations gives you more resolution than a single broad label. But more populations also means more places for the algorithm to split slightly different ancestry between similar groups.
A reference population is not an ancestor
The common mistake is treating a reference population as if it were your actual relatives. A present-day reference group is a proxy. It is a sample of living people, usually selected because their recent ancestors came from one place. That does not mean your ancestors were those people. It means your DNA resembles them more than the other groups in the panel.
Ancient reference groups go further. They are built from published ancient genomes, sometimes from a single burial site or a small set of individuals. The label Corded Ware or Yamnaya is a convenience. It shows shared genetic drift with that set, not a direct line of descent. Ancient matches are interesting, but they are not a family tree.
Autosomal DNA cannot settle a lineage or identify a tribe by itself. Broad patterns are signal; a named tribe or ancestor is usually beyond the resolution.
Why do my ancestry percentages change between companies?
Because the reference panel changes, and the labels change with it.
Different companies recruit different reference populations, even when they use the same language. One company might have a reference group called Scandinavia made up of samples from Denmark, Sweden and Norway. Another might split that into separate Danish, Swedish and Norwegian panels. Your DNA has not changed. The comparison set has.
The algorithm changes too. Some tests use more aggressive smoothing, forcing nearby chunks into the same group. Others keep more small segments. That affects the percentages. The same raw file can produce different numbers at different services, and both can be internally valid.
What matters is whether the broad pattern holds. If you are half West African and half Northern European, that should show up across any reasonable panel. A 4 percent difference between England and Northwestern Europe and Germanic Europe is noise, not a contradiction.
How to read a result without treating it as fact
Use the label as a hypothesis, not a verdict.
- Treat percentages as ranges. A 23 percent result might reasonably be a few points higher or lower with a different panel.
- Look for named panels. A panel with 15 named regions and 157 reference populations is easier to judge than a single White or European label.
- Separate region percentages from ancient matches. They answer different questions.
- If you already have a raw file, run it through a second reader to see whether the main continental pattern stays the same.
What you see on a graph is not one measurement. It is many small assignments summed into a percentage. Read it the way you would read a weather model, not a birth certificate.
If you already have a raw DNA file, the free reader is where you can see it in your own numbers. If you have no file yet, the HealthOSX DNA Kit is the kit with the 15-region, 157-reference-population panel described here.
New to raw DNA files? Start with the full guide: 23andMe Raw Data Interpretation Free in Your Browser
Written by
HealthOSX Research