Biological age tests, most commonly epigenetic clocks, estimate how "aged" your cells appear based on chemical patterns in your DNA, distinct from your chronological age, the number of years since you were born. The core idea is that biological aging happens at somewhat different rates in different people, and a test that can detect this variation directly, rather than inferring it indirectly from symptoms or a handful of biomarkers, could offer a genuinely useful longevity signal, though the field is younger and less standardized than most of the other testing methods covered in this encyclopedia, worth knowing upfront before treating any single result as a settled, precise fact about your own aging trajectory.
The conceptual distinction between chronological and biological age is covered in the chronological versus biological age entry. This page covers the tests themselves.
- Epigenetic clocks estimate age from DNA methylation patterns, chemical markers on DNA that shift predictably as cells age.
- GrimAge, a newer clock, is widely considered the strongest current predictor of mortality and disease risk among published clocks.
- Different clocks can give meaningfully different results for the same person, a genuine reliability limitation worth knowing before testing.
- No epigenetic clock has FDA approval as a diagnostic tool, results are best treated as directional, not precise.
How epigenetic clocks work
DNA methylation is a chemical process where small molecules called methyl groups attach to specific locations on your DNA. They influence how genes are expressed without changing the genetic code itself, forming part of the epigenome, the layer of chemical modification sitting on top of your genes.
Methylation at many of those locations changes in consistent, predictable ways as cells age. Consistent enough that researchers can build statistical models predicting chronological age from methylation data alone, with striking accuracy in the populations the models were trained on.
Some age-related sites track chronological age tightly across nearly everyone studied. Others vary more with actual biological ageing rate, and that is where the health-relevant signal potentially lives.
An epigenetic clock is that kind of trained model. Built by analysing methylation from thousands of people of known age, then identifying the DNA locations whose methylation most reliably tracks with age.
Once built, it can be applied to a new person's sample, usually blood or saliva, to produce a predicted epigenetic age. In a healthy person that tends to track chronological age fairly closely, and some research suggests it can diverge with lifestyle, disease burden and other factors.
The gap between the two, sometimes called age acceleration, is usually the more interesting number than the raw epigenetic age.
The measurement technology is either methylation microarrays or targeted sequencing, reading methylation status at hundreds to thousands of genomic locations. It has become much cheaper over the past decade.
That is most of why consumer biological age testing exists commercially at all. The lab work used to be too expensive and too slow for anything outside dedicated research settings.
Not the same as “metabolic age”
Bathroom scales and fitness trackers often report a “metabolic age”, and it is a different measurement entirely. That figure is usually derived from your body composition and estimated resting metabolic rate, compared against population averages for your age.
An epigenetic clock is reading chemical marks on your DNA. Different input, different method, different claim. The two can disagree sharply on the same person, and neither is a correction of the other.
Worth knowing which one a product is selling you, because the word “age” is doing a lot of work in both and only one of them involves your cells.
The major epigenetic clocks
Epigenetic clocks fall into three generations, and the generation explains most of the disagreement between them.
First generation clocks were trained to predict calendar age. That is also their limitation: an algorithm optimized to reproduce your birthday is optimized away from telling you anything your birthday doesn't. Second generation clocks were trained on clinical markers and mortality instead, which is the shift that made deviation from chronological age mean something. Third generation estimates the rate you are currently ageing rather than a cumulative total.
The Horvath clock, published in 2013 and built from 353 methylation sites, was among the first widely used multi-tissue clocks. It worked reasonably well across many tissue types rather than being specific to blood or skin, which was a genuine methodological achievement at the time.
The Hannum clock arrived the same year, trained on blood specifically. Optimised for one tissue rather than the breadth Horvath aimed at.
Both were genuine breakthroughs, and both were built to predict chronological age. That turns out to be a different goal from predicting health outcomes.
Someone can have methylation that predicts their birthday almost exactly, while the same clock says little about how long they will live or how healthy they are now.
PhenoAge and GrimAge are the second generation. They were trained on health outcomes and mortality directly, using methylation correlated with clinical biomarkers and, for GrimAge, mortality data itself.
That change in training target is most of why they outperform the first generation in outcome research. What a model is optimised to predict shapes what it turns out to be good at.
DunedinPACE, published in 2022, is a third generation and a different kind of measure. First- and second-generation clocks report a cumulative age, an estimate of how old you appear. DunedinPACE reports a rate: how fast you are currently ageing, expressed as biological years per chronological year.
A pace measure is conceptually the more useful one for tracking whether an intervention is working, because it reports the current rate rather than a lifetime total that recent changes barely move.
What a randomised trial actually showed
Most claims that biological age can be lowered rest on observational data. The CALERIE trial is the exception, and what it found is more interesting than a simple yes.
220 adults without obesity were randomised to 25% caloric restriction or an ad libitum diet for two years. The intervention slowed DunedinPACE by roughly 2 to 3%, with a Cohen's d of 0.3 at 12 months and 0.2 at 24 months, p<0.01 for both.
It produced no significant change in PhenoAge or GrimAge, and effects on some other clocks ran in the opposite direction.
That split is the point. The pace-of-ageing measure moved. The static cumulative-age clocks did not. Which is what the three-generation framing predicts: a clock reporting a rate should respond to an intervention on a two-year timescale, and one reporting a cumulative total mostly should not.
On whether 2 to 3% matters, the authors note that in an independent study of older adults, 3% slower DunedinPACE was associated with a 15% lower risk of death. They are also explicit that the treatment effect sizes were small and that a conclusive test needs long-term follow-up on hard endpoints.
Comparing the epigenetic clocks
| Clock | Trained to predict | Strength as a predictor |
|---|---|---|
| Horvath (2013) | Chronological age | Modest |
| Hannum (2013) | Chronological age (blood) | Modest |
| PhenoAge (2018) | Clinical biomarker-based "phenotypic age" | Strong |
| GrimAge (2019) | Mortality & disease risk directly | Strongest |
| DunedinPACE (2022) | Pace of aging, as a rate rather than a total | Strong, and designed for tracking change |
Why GrimAge stands out
GrimAge was built differently from its predecessors. Rather than training a model to predict chronological age and hoping that also predicted health, researchers trained it on methylation correlated with known mortality risk factors, including estimated smoking history and several clinical biomarkers.
Critically, they then validated it directly against mortality data from long-term cohorts. That design is why GrimAge consistently outperforms earlier clocks in head-to-head comparisons predicting all-cause mortality, cardiovascular disease and cancer risk.
That design has a practical consequence worth knowing before you test. Because GrimAge incorporates a methylation surrogate for smoking history, a former smoker will see a large age acceleration driven substantially by that history. It is real signal, and it is not something current behaviour can move much, which matters if you are reading the result as a scorecard for what you are doing now.
None of which makes GrimAge finished. It remains under active refinement, with successor models including GrimAge2 building on the same approach.
Among published options available now, though, it carries the strongest evidence base for anyone who wants a biological age result with a real link to health outcomes rather than an interesting estimate of their birthday.
One caveat runs under all of it. This outcome research comes from population cohorts, showing the clocks predict group-level risk reasonably well. That is a different claim from predicting any one person's trajectory, and the distinction tends to disappear in how these tests are marketed.
How to get an epigenetic age test
Several direct-to-consumer companies offer epigenetic age testing, usually via a blood draw or an at-home saliva or blood-spot kit sent away for methylation analysis.
Pricing and methodology vary a lot. The thing worth checking is which clock or clocks a provider actually reports, because as the comparison above shows they are not equivalent, and a service reporting only a first-generation clock is selling you a good estimate of your birthday.
Results usually arrive as a single biological age number, sometimes next to your chronological age, and increasingly with a breakdown across several clocks run on the same sample. That breakdown is useful, because it shows how much the models disagree about you specifically.
Some services track change across repeated tests, which helps for spotting a trend. The retest variability discussed below applies to any single change between two tests, though.
How reliable epigenetic clock results are
The clearest way to see the problem is to measure the same sample twice. Across six prominent epigenetic clocks, technical noise alone produced deviations of up to 9 years between replicates. Not two people, not two timepoints. The same sample, measured twice.
Put concretely: a clock that reports you as biologically 50 on one test and 59 on the next has told you nothing about your ageing, and everything about its own measurement error.
There is a published partial fix. Higgins-Chen and colleagues retrained those six clocks using principal components, and their PC-based versions brought most replicates into agreement within about 1 to 1.5 years. If a test names the clock it uses, whether it is a PC version is a reasonable thing to ask.
A further distinction is emerging that the reliability figures above don't capture. Technical reproducibility, meaning the same sample measured repeatedly, is a different question from biological reliability, meaning the same person sampled repeatedly within hours, before and after meals, or under acute stress. Recent work reports that most clocks achieve good technical reproducibility but only moderate biological stability, that PCGrimAge was the only clock reaching an intraclass correlation above 0.75 for biological reliability, and that technical reproducibility did not predict biological reliability. A clock can be highly repeatable and still move around inside the same person for reasons unrelated to ageing. That analysis is a preprint and has not been peer-reviewed, so treat it as a direction of travel rather than a settled finding.
Different clocks run on the exact same sample from the same person can produce meaningfully different estimates, because each was trained on different data with different assumptions about which methylation sites matter.
That is not a technical footnote. It is an actively discussed limitation, and it is why a single number from one commercial test should not carry more confidence than the science currently supports, particularly when comparing across companies or clock versions.
No epigenetic clock currently has FDA approval or equivalent clearance as a diagnostic tool. These are wellness or research-oriented products, and they should not drive significant medical decisions without independent clinical evaluation.
Test-retest reliability also varies by provider, and has not been independently validated to the degree it has for established markers like HbA1c or a standard lipid panel.
Sample handling adds variability of its own. Methylation testing is sensitive to how a sample is collected, stored and processed, and labs' internal quality control standards are not uniformly disclosed or independently audited the way clinical testing for established biomarkers is.
Common misconceptions
"My biological age number is a precise, scientifically settled measurement." It is a statistical estimate from an evolving field, with meaningful variation between clocks. Treat it as a rough directional signal. Blood pressure has decades of standardised protocol behind it; epigenetic testing does not have that yet.
"A lower biological age than my chronological age proves my lifestyle choices are working." It is a reasonable data point. The research linking specific interventions to reliably lowered epigenetic age in an individual is still developing, and a population-level correlation with healthy behaviour does not guarantee the same causal relationship in your result, particularly given the variability above.
"All biological age tests measure the same thing the same way." Different clocks were built with different goals and different training data, and produce meaningfully different results. Checking which clock a service uses matters more than treating "biological age testing" as one standardised category the way a lipid panel is.
"Biological age testing is essentially the same as other biomarker testing in this encyclopedia." Most other markers covered here, VO2 Max, blood pressure, HbA1c, have decades of standardized measurement and large-scale outcome research behind them. Epigenetic clocks are a genuinely newer, less standardized field, worth engaging with as an interesting emerging area rather than treating with the same confidence as more established, longer-validated testing methods.
How much weight to give a biological age result
The usual framing treats a biological age result as the single number that matters most. It is an interesting summary signal with real reliability limitations, and it sits well behind directly measured markers.
The major clocks are built differently and can disagree with each other on the same sample. That disagreement is the reason to hold the result loosely rather than build a plan around it.
If the number is worse than you expected, the useful response is to look at the measured markers underneath rather than to retest the clock.
The app prioritizes directly-measured, outcome-validated markers first, with biological age testing as an optional, later addition, not the foundation of your score.
The tests are not cheap either, which sharpens the question. If the result rarely changes what you would do, and the clocks can disagree on the same sample, the markers underneath are the better purchase. Blood pressure, aerobic fitness, strength and glucose control cost far less to measure and each points at something you can act on.
Sources
Key references for the claims on this page. Where a figure is attributed to a specific study or body, it is named here.
- Horvath S. DNA methylation age of human tissues and cell types. Genome Biology, 2013;14(10):R115. The multi-tissue clock built from 353 methylation sites. PMID 24138928
- Hannum G, et al. Genome-wide methylation profiles reveal quantitative views of human aging rates. Molecular Cell, 2013. PMID 23177740
- Levine ME, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging, 2018. PhenoAge. PMID 29676998
- Lu AT, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging, 2019. PMID 30669119
- Belsky DW, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife, 2022. PMID 35029144
- Higgins-Chen AT, Thrush KL, Wang Y, et al. A computational solution for bolstering reliability of epigenetic clocks: implications for clinical trials and longitudinal tracking. Nature Aging, 2022. Source of the up-to-9-year replicate deviation and the PC-clock figures. DOI
- Waziry R, Ryan CP, Corcoran DL, et al. Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial. Nature Aging, 2023. n=220, two years, randomised. Source of the DunedinPACE finding. PMID 37118425
Frequently asked
What is an epigenetic clock?
An algorithm estimating biological age from DNA methylation patterns, chemical markers on DNA that change predictably as cells age.
Are biological age tests reliable?
They're a legitimate research area with genuine predictive value at a population level, but different clocks give different individual results, and none has FDA approval as a diagnostic tool.
Which clock is considered the best?
GrimAge is widely considered the strongest current predictor of mortality and disease risk, outperforming earlier clocks in most published comparisons.
Can I lower my biological age?
Population research links healthy lifestyle factors, exercise, diet, sleep, and not smoking, to lower epigenetic age, though individual-level causal evidence for specific interventions is still developing.
How often should I retest?
Given current test-retest reliability limitations, annual testing at most is reasonable, more frequent testing is unlikely to reveal a meaningful trend distinct from normal measurement noise.