Why Two Biological Age Tests Can Give You Different Answers

Two biological age tests can disagree because most of them measure different things, are calibrated against different reference populations, and carry more measurement noise than their tidy single-number output suggests. A first-generation epigenetic clock estimates how old your cells look compared with a training set, while a pace-of-aging measure estimates how fast you are aging right now, so they answer different questions. Add the technical noise built into DNA methylation arrays, and the same tube of blood can yield estimates that differ by years. These scores were designed as research tools for comparing groups, not as personal verdicts.

What the tests actually measure

The oldest and most familiar tests are the first-generation epigenetic clocks, published by Steve Horvath and by Gregory Hannum in 2013. They read DNA methylation, the chemical tags that sit on top of your genome, at a few hundred sites and combine them into an estimate of chronological age. Horvath's model uses 353 sites; Hannum's uses 71. These clocks were trained to predict how many birthdays a person has had, and they do that reasonably well across a population.

A pace-of-aging measure works differently. DunedinPACE, described by Daniel Belsky and colleagues in eLife in 2022, was trained on the Dunedin cohort, a group of roughly a thousand people in New Zealand tracked with repeated organ-function measurements across two decades. Instead of guessing your age, it estimates the rate at which your body is deteriorating, expressed as biological years accrued per calendar year. A useful way to hold the distinction: a chronological-age clock is an odometer reading total mileage, while DunedinPACE is a speedometer reading current speed. An odometer and a speedometer will never show the same value, and neither is broken.

Calibration is different too

Every clock is anchored to the population it was trained on. Horvath's original clock drew on a mix of tissues; Hannum's was built from adult blood; DunedinPACE was calibrated to one deeply studied New Zealand birth cohort. When your sample is scored, it is being compared with that reference group, and different reference groups sit at different baselines. A result that reads as older than your age against one training set can read differently against another, before any biology enters the picture. Direct-to-consumer companies rarely publish which clock version and which reference distribution sit behind the number they mail you, which makes cross-brand comparison close to meaningless.

The noise problem

This is the reason two runs of the same test, on the same sample, can disagree. In 2022, Albert Higgins-Chen and colleagues showed in Nature Aging that technical noise in the methylation measurement could shift a person's estimated epigenetic age by as much as nine years between replicate readings of a single sample. Nine years is not biology. It is the assay. Their fix was to rebuild the clocks using principal component analysis, which pools information across many methylation sites rather than trusting individual ones, and that brought most replicate pairs into agreement within about 1.5 years. DunedinPACE was designed with the same lesson in mind, keeping only the most reliable probes, and its developers reported test-retest reliability around 0.96 on a scale where 1.0 is perfect. A test that has not been engineered for reliability can hand you a different number on Tuesday than it did on Monday, with no change in your health.

Why direct-to-consumer scores drift

A December 2025 review in Epigenomics by Abner Apsley, Idan Shalev, and colleagues laid out why these tools struggle at the individual level even when the underlying science is sound. Methylation estimates are sensitive to which array platform a lab uses and to small changes in the data-processing pipeline, so a company that updates its equipment or its normalization method can move your score without moving you. Methylation itself fluctuates over minutes to hours, and recent illness, stress, or a shift in blood cell composition can nudge the reading. The authors noted that median errors in these clocks commonly run at or above 3.6 years, and concluded that the clocks, as currently built, fail common standards for clinical utility as personal diagnostics.

None of this means the science is fake. It means the tests were validated to compare groups, not to grade one person, and two companies using different clocks, different reference populations, and different pipelines have little reason to converge on the same answer for you.

How to read your own result

Treat a single biological age score as one noisy data point, not a diagnosis. If you want to learn anything from these tests, the most defensible approach is to use the same test, from the same company, on the same platform, over time, and to watch the direction of the trend rather than any single value. A measure engineered for reliability and tracked longitudinally can hint at whether your trajectory is bending, which is closer to what pace-of-aging measures were designed to detect. A one-time score compared against a friend's one-time score from a different vendor tells you very little.

The ordinary levers still carry far more evidence than any test result: sleep, physical activity, not smoking, and managing blood pressure and metabolic health. This article is educational and is not medical advice; decisions about your health belong in a conversation with a qualified clinician who knows your history.

Biological age testing is a genuinely promising research field, and the second-generation measures are meaningfully better than the first. The fair reading of the current evidence is that these scores are instruments for studying populations, and a disagreement between two of them usually reflects how they were built rather than a hidden truth about your body.