Can DNA Testing Be Wrong? Reasons for Inaccurate Results

DNA testing can absolutely be wrong, and the reasons range from mundane lab mix-ups to genuinely strange biological phenomena that most people never hear about. The error rate depends heavily on the type of test: a forensic crime-scene profile, a direct-to-consumer health report, and a prenatal screening test each fail in different ways and for different reasons. One study found that roughly 40% of disease-risk variants flagged in raw data from consumer genetic tests turned out to be false positives when checked with clinical-grade methods. That number alone should make anyone cautious about treating a DNA result as absolute truth.

Contamination and Human Error in the Lab

The single biggest source of DNA testing errors is also the most boring: something goes wrong in the lab. Contamination, where DNA from an unintended source gets mixed into a sample, is a persistent problem across every type of DNA analysis. In sequencing studies, even small amounts of contamination can cause the wrong genetic variants to be called, leading to false matches or false disease associations.¹ Genotyping arrays, which are what most consumer DNA kits use, are highly accurate in ideal conditions but sensitive to sample mixing. When DNA from two people accidentally gets combined before or during processing, the resulting errors can produce false positives that look convincing.²

In forensic settings, the stakes are higher and the problem is even more studied. An analysis of forensic DNA laboratory errors found that contamination and human error were the most common causes of failures. Human errors, like mislabeling a tube or entering data incorrectly, could usually be caught and corrected. But gross contamination of crime-scene samples often caused irreversible damage, making it the most significant error source in forensic DNA work.³ These are not hypothetical concerns. They are documented failures that have affected real cases.

Why Consumer Genetic Tests Have Such High False Positive Rates

If you’ve spit into a tube and mailed it to a company for health-related genetic insights, the results deserve more skepticism than most people give them. When researchers took the raw variant data that direct-to-consumer companies reported to customers and verified those variants using clinical-grade sequencing, they found that about 40% of variants flagged in the raw data were false positives.⁴ That means if your consumer test flagged five disease-risk variants, roughly two of them might not actually be in your DNA at all.

Part of the problem is what the sample is made of. Most consumer kits collect saliva or buccal (cheek) swabs rather than blood. These are convenient, but saliva is teeming with oral bacteria. When the whole sample goes through sequencing, bacterial DNA can reduce the accuracy of variant detection compared to blood-derived samples.⁵ Clinical labs drawing blood for a diagnostic test start with a cleaner sample, which is one reason clinical results tend to be more reliable than consumer ones.

Another technical issue is something called allele dropout. During the process of copying tiny amounts of DNA for analysis, one copy of a gene can fail to get amplified while the other copy works fine. This makes a person who carries two different versions of a gene look like they carry only one, which can lead to a missed diagnosis or a wrong one. Allele dropout tends to happen when natural genetic variation in a person’s DNA interferes with the chemical primers used to start the copying process.⁶ It is a known risk for any test that relies on amplification, from consumer kits to clinical diagnostics.⁷

Sequencing Technology Has Blind Spots

Even when everything in the lab goes perfectly, the technology itself has inherent limitations. Most DNA tests today use short-read sequencing, which chops your genome into tiny fragments, reads each one, and then computationally stitches them back together by comparing them to a reference genome. This works well for detecting small, straightforward changes like a single-letter swap in your DNA. It works poorly for detecting large structural changes, such as big deletions, duplications, or rearrangements, which account for a meaningful chunk of the genetic variation that causes disease.⁸

Newer long-read technologies handle some of these structural variants better, but they have their own blind spots. A study comparing short-read and long-read approaches found that long-read assembly methods missed the vast majority of large copy-number variants that short-read depth-based methods could detect. On average, each genome had about 167 large copy-number variants that only depth-based methods caught, and nearly 90% of those were invisible to long-read assembly alone.⁹ No single sequencing approach catches everything.

The reference genome itself also matters. For years, the human reference genome had gaps and errors that caused false variant calls. A more complete version of the reference, published in 2022, eliminated tens of thousands of false positive variants per sample and reduced false positives in hundreds of medically relevant genes by up to twelve-fold.¹⁰ That means many DNA tests performed before this improved reference existed were calling variants that were never actually there.

Another source of confusion is pseudogenes, which are defunct copies of real genes scattered throughout the genome. Because pseudogenes look very similar to the working genes they were copied from, sequencing reads can get assigned to the wrong location. This generates false positive variants in the real gene, or causes real mutations to be missed entirely.¹¹ A related problem occurs with mitochondrial DNA analysis, where short fragments of mitochondrial DNA that have been copied into the nuclear genome over evolutionary time can confuse sequencing pipelines, biasing variant calls toward the reference sequence and masking real mutations.¹²

Forensic DNA and the Problem of Transfer

In criminal investigations, a DNA match at a crime scene is often treated as near-conclusive evidence. But DNA does not stay where it is deposited. Secondary DNA transfer, where your DNA moves from you to an object you never touched via an intermediary person or surface, is a well-documented phenomenon. In one controlled study, researchers found that secondary transfer was detectable in 85% of samples. In five of those samples, the person whose DNA had been transferred indirectly was either the only contributor or the major contributor on a knife they had never touched.¹³ Even the nitrile gloves worn by crime-scene investigators can act as vectors, picking up DNA from one piece of evidence and depositing it on another.¹⁴

When crime-scene DNA comes from multiple people mixed together, the analysis gets even more uncertain. Modern forensic labs use probabilistic genotyping software to estimate the likelihood that a specific person contributed to a mixture. But these programs can produce wildly different answers from the same evidence. In one federal case, two widely used programs evaluated the same DNA mixture sample. One program returned a likelihood ratio of 24 in favor of the person not being a contributor, essentially saying the DNA probably was not theirs. The other program returned a likelihood ratio of up to 16.7 million in the opposite direction, strongly implicating the person. The difference traced back to subtle variations in modeling assumptions and analytical thresholds between the two programs.¹⁵

Population genetics also plays a role in forensic accuracy. A study examining DNA mixture analysis across different population groups found that accuracy decreased for groups with lower genetic diversity. The false positive rate went up when contributors to a mixture came from populations where individuals share more genetic markers in common, making it harder for the software to distinguish one person from another.¹⁶

When Your Body Has More Than One Genome

Some of the most dramatic DNA testing errors come not from the lab or the technology but from the person being tested. A small number of people are chimeras, meaning their body contains cells with two genetically distinct sets of DNA. This happens naturally when two fraternal twin embryos fuse very early in development, producing a single person whose blood cells might have one genome while their skin or reproductive cells have another. Chimerism can also be acquired through blood transfusions or organ transplants.¹⁷

The consequences for DNA testing can be bizarre. If a man with chimerism conceives a child using sperm that carries a different genome than his blood, a standard paternity test drawn from blood will falsely exclude him as the father.¹⁸ There are documented cases where a child’s blood typing showed results seemingly impossible for the declared parents, and only extensive molecular testing revealed that the child was a chimera carrying two distinct sets of genetic markers at multiple locations across the genome.¹⁹

Mosaicism, which is related but more common, occurs when a single person’s cells acquire different mutations during development. Unlike chimerism, which involves two separate genomes, mosaicism means cells in different tissues have small differences from a shared starting genome. This can cause a test run on blood to miss a variant that is present in other tissues. In a study comparing saliva and blood, researchers found 20 individuals with a mosaic abnormality visible in saliva but completely absent from blood.²⁰ Depending on which tissue you test, you may get a completely different answer about what genetic variants a person carries.

The Identical Twin Problem

Standard forensic DNA profiling cannot tell identical twins apart. The short tandem repeat markers used in conventional testing are the same in both twins because they come from the same fertilized egg. This has created genuine problems in criminal cases and paternity disputes where both twins are potential suspects or fathers.²¹

Identical twins do accumulate a small number of genetic differences after the embryo splits, but finding those differences requires ultra-deep sequencing at a level far beyond what routine forensic testing provides. Researchers have demonstrated that this approach can work in real cases, identifying rare mutations unique to one twin.²² But the technique remains expensive, time-consuming, and far from standard practice.²³

Prenatal Screening Is Not Diagnostic

Non-invasive prenatal testing, commonly called NIPT, analyzes fragments of fetal DNA circulating in the pregnant person’s blood to screen for chromosomal conditions like Down syndrome. It is highly sensitive, but it is a screening test, not a diagnostic one, and the distinction matters enormously. NIPT analyzes DNA that comes primarily from the placenta, not directly from the fetus. If the placenta has a chromosomal abnormality that the fetus does not, or vice versa, the test gives a wrong answer.²⁴

The most common cause of false results on NIPT is confined placental mosaicism, where the placenta carries a chromosomal change that is not present in the fetus.²⁵ False negatives also occur: one study documented a case where a normal NIPT result was followed by a 20-week ultrasound that revealed trisomy 18 the test had missed. The same study found that about 24% of chromosome abnormalities in high-risk pregnancies involve conditions that NIPT does not currently screen for at all.²⁶

There is also a less well-known source of false NIPT results: maternal cancer. Tumors shed DNA into the bloodstream, and that tumor DNA gets picked up by NIPT along with the placental DNA. An abnormal NIPT result that does not match the fetal karyotype has, in some cases, turned out to be the first sign that the pregnant person had an undiagnosed malignancy.²⁷

A Variant Called “Pathogenic” Today Might Not Be Tomorrow

Even when the lab work is flawless and the sequencing is accurate, the interpretation of results can change. Genetic variants are classified on a scale from benign to pathogenic based on the evidence available at the time. As researchers learn more, those classifications get updated. Studies have found that variant reclassification rates range from about 4% to nearly 59%, depending on the gene, the lab, and how much time has passed since the initial call.²⁸ A variant that triggered a cancer-screening protocol two years ago might be downgraded to benign once more population data accumulates.

The problem runs in both directions. An examination of pathogenic variant entries in ClinVar, a major public database used by clinical labs, found that about 65% of downgraded variants had been classified as pathogenic because the published evidence simply was not strong enough to support that call. In roughly 10% of cases, the original research paper actually described the variant correctly as a benign polymorphism or mere risk factor, but someone submitted it to the database as pathogenic or likely pathogenic anyway.²⁹ So even a correctly sequenced and correctly detected variant can lead to a wrong medical conclusion if the interpretation in the database is outdated or was wrong from the start.

Ancestry Estimates and Population Bias

Ancestry testing is the application where people are most likely to notice inconsistencies firsthand, because it’s common for friends or family members to test with different companies and compare results. The percentages you get back are estimates based on a reference database of genotypes that each company has built from its own customers and from public sequencing projects. Different companies use different reference populations, different algorithms, and different geographic labels. When researchers tested identical twins across multiple companies, the twins sometimes received noticeably different ancestry breakdowns from the same company, and results varied considerably across companies.³⁰

The underlying issue extends well beyond ancestry entertainment. Polygenic risk scores, which estimate your genetic likelihood of developing conditions like heart disease or diabetes based on the combined effects of many small genetic variants, have a well-documented accuracy problem across populations. These scores are typically developed using data from people of European descent, and their predictive power drops substantially when applied to people of other backgrounds.³¹ A person of West African or East Asian ancestry could receive a risk score that is misleadingly high or misleadingly low, not because their DNA was read wrong, but because the statistical model behind the score was built on a population that does not represent them. This is not a future concern to be addressed someday. It is a current limitation baked into widely used tools.

When Different Platforms Disagree on the Same Sample

One underappreciated source of error is that different testing platforms sometimes give different answers when analyzing the same biological sample. In preimplantation genetic testing of embryos, researchers compared results from two major platforms analyzing the same 105 embryos flagged as mosaic. The two platforms fully agreed on only about 48% of cases. They partially agreed on another 18%. And they completely disagreed, giving opposite or unrelated diagnoses, on more than a third of the samples. Some embryos called mosaic by one platform were called completely normal by the other.³²

This kind of inter-platform disagreement matters whenever clinical decisions hinge on the result. In the IVF context, an embryo diagnosed as mosaic might be set aside rather than transferred, potentially discarding a healthy embryo. In clinical genetics, a variant detected by one sequencing approach but not another could mean the difference between a diagnosis and years of uncertainty. The technology is powerful, but treating any single platform’s output as ground truth, particularly for borderline or low-level findings, is a mistake. Confirmation with an independent method remains the gold standard precisely because no single technology is immune to its own idiosyncrasies.