Ask a laboratory how accurate its paternity test is and you will get a number like 99.99%. Ask what the number means and the conversation usually stops. It is worth understanding, because the figure is not a confidence rating in the everyday sense, and the situations that actually threaten a result are not the ones people worry about.
Two answers, not one
A paternity test produces one of two outcomes, and they have very different accuracy profiles.
Exclusion is essentially absolute. If the tested man does not carry the DNA the child must have inherited from the biological father, he is not the father. Labs require mismatches at two or more markers before calling an exclusion, precisely so that a single mutation cannot produce a false negative. When a report says 0% probability of paternity, that is as close to certainty as testing gets.
Inclusion is a probability. If the man matches at every marker, the lab cannot literally prove he is the father, because it has not tested every man on earth. Instead it calculates how much more likely the evidence is if he is the father than if a random unrelated man is. That ratio becomes the probability of paternity.
Decoding the number
A probability of paternity of 99.99% corresponds to a combined paternity index of about 10,000 to one. In plain terms: the genetic evidence is roughly ten thousand times more consistent with this man being the father than with an unrelated man from the same population being the father.
| Combined paternity index | Probability of paternity |
|---|---|
| 100 : 1 | 99.01% |
| 1,000 : 1 | 99.90% |
| 10,000 : 1 | 99.99% |
| 1,000,000 : 1 | 99.9999% |
Most US states set 99% as the statutory threshold at which a genetic test creates a legal presumption of paternity. Modern tests clear that bar by orders of magnitude. Testing more markers pushes the index higher, which is why labs that once used 16 markers now commonly run 20 to 24 or more.
Why no report ever says 100%
A test tells you that the man's DNA is consistent with fatherhood. It cannot tell you that no other man's DNA would also be consistent. Population genetics leaves a residual, and honest labs report it. Anyone advertising a 100% inclusion is either rounding or selling something.
The things that genuinely reduce accuracy
Untested close relatives
This is the real one. Brothers share about half their DNA. A father and son share half. If two brothers could both be the father and only one is tested, the tested brother can return a very high probability of paternity and still be wrong. Identical twins cannot be distinguished by standard STR testing at all.
Every reputable lab asks about this on the intake form. Answer it honestly. If a close relative is in the picture, test him too - it costs a fraction of what a wrong answer costs.
Mutations
STR repeat counts occasionally change when they are passed on. A single-marker mismatch in an otherwise perfect profile usually indicates a mutation, not an exclusion. Labs respond by expanding to additional markers and adjusting the calculation. This is a routine, handled situation - but it is also why a lab that tests only a handful of markers is a worse lab.
Sample problems
Contaminated, degraded, or mixed samples do not usually create a wrong answer; they create a failed one. A swab used within 30 minutes of eating, or left damp in a sealed bag, may not yield a clean profile. The lab asks for a recollection.
Human error and substitution
This is where accuracy is actually lost in practice, and it is not a laboratory problem. If the wrong person swabs, or someone swaps envelopes, the lab produces a perfectly accurate report about the wrong people. Chain of custody exists entirely to close this gap, and it is the reason courts will not accept a home-collected sample. See our guide to chain of custody.
Does an at-home kit test less accurately?
No. This is the most common misconception in the field. The kit you buy off a shelf is generally processed by the same accredited laboratory, on the same instruments, using the same marker panel as a $450 legal test. What you lose is not analytical accuracy - it is proof of identity and admissibility.
Where at-home kits do underperform is in failure rate, because collection is unsupervised. Get the swab right and the science is the same.
How accreditation protects the number
The probability on the report is only as good as the process behind it. Two credentials matter in the US:
- AABB accreditation for relationship testing, which is what courts, agencies, and immigration authorities look for. The AABB publishes its list of accredited facilities.
- CLIA certification, the federal standard for clinical laboratory operations, administered under the Clinical Laboratory Improvement Amendments.
A lab with neither is not necessarily wrong, but you have no independent basis to believe it is right.
Prenatal and kinship tests are a different conversation
Non-invasive prenatal paternity testing reaches comparable inclusion probabilities but depends on there being enough fetal DNA in the mother's blood, which is why it has a gestational age minimum and a redraw rate. Sibling, grandparent, and avuncular tests are inherently weaker: they compare shared ancestry rather than direct inheritance, and typically produce likelihood ratios rather than 99.99% answers. Testing more relatives improves them.
For general background on how genetic tests are validated and reported, MedlinePlus has a plain-language overview, and the FDA maintains guidance on direct-to-consumer tests.
The practical summary
- An exclusion is definitive. Trust it.
- An inclusion at 99.9% or higher is about as strong as evidence gets - unless a close relative was never tested.
- Analytical accuracy does not depend on price or on whether the test is legal or at-home.
- What you pay extra for in a legal test is proof of who was swabbed, not a better answer.