Best Epigenetic Clock: GrimAge vs PhenoAge vs DunedinPACE

The one-sentence version

There is no single "best" epigenetic clock, because the leading clocks were built to do different jobs: GrimAge is the strongest at predicting time-to-death, PhenoAge is tuned to current disease and physiological dysfunction, and DunedinPACE measures a rate of aging rather than an age. Asking which is best is like asking whether a thermometer or a speedometer is the better instrument — it depends entirely on what you're trying to read. And underneath all of them sits a limit that no clock has escaped: every one of these tools is validated to rank-order populations, and none has been shown to be a target you can move to extend your own life. For where biological-age testing fits in the wider toolkit, start with our explainer on whether epigenetic clocks actually work and our pillar on longevity medicine: what's proven vs hyped.

A clock is just a weighted sum of methylation sites

Every clock here reads the same raw material: DNA methylation, the pattern of chemical tags on your DNA that shifts predictably with age. The differences come from what each clock was trained to predict. The first-generation Horvath clock (2013) was trained simply to estimate chronological age across tissues — an impressive technical feat, but one that, by design, treats deviating from your birthday as the signal¹. The clocks people actually argue about today are the "second-generation" ones, trained not on age but on health and survival outcomes.

GrimAge: the mortality specialist

If you want the single clock with the strongest record at predicting who dies sooner, it's GrimAge. Rather than training on age, GrimAge was built to predict lifespan and healthspan: it combines DNA-methylation surrogates for smoking-pack-years and seven plasma proteins into a composite that, in its founding paper, strongly predicted time-to-death and time-to-disease across multiple cohorts². A second iteration, GrimAge version 2, refined those surrogates and improved prediction further, particularly in some populations the original handled less well³. The honest framing: GrimAge is the best-evidenced risk stratifier of the group. What it is not is a proven dial — no trial has shown that lowering your GrimAge makes you personally outlive your original prediction.

PhenoAge: the disease-and-dysfunction clock

PhenoAge (Levine 2018) was trained on a "phenotypic age" derived from nine clinical chemistry markers — things like albumin, creatinine, glucose, and C-reactive protein — that together track current physiological dysfunction⁴. The result is a clock that leans toward present disease burden and morbidity rather than the pure mortality-forecasting that GrimAge specializes in. In practice GrimAge and PhenoAge often agree directionally, but PhenoAge's appeal is that it's anchored to ordinary blood-panel biology — which is also why a simplified version of the underlying phenotypic-age formula can be computed from a basic blood draw, no methylation array required (we cover that in free biological-age tests and calculators).

DunedinPACE: the speedometer, not the odometer

DunedinPACE (Belsky 2022) is conceptually different from both. It doesn't estimate an age at all — it estimates a pace, calibrated against two decades of longitudinal organ-system decline in a single birth cohort followed since 1972. A reading of 1.0 means you're aging one biological year per chronological year; 1.2 means roughly 20% faster⁵. That design makes it the most appealing clock for tracking change over time, because a rate is exactly what you'd want to watch if you were testing an intervention.

DunedinPACE has two further things going for it. First, it was developed in the reliability-conscious lineage that emerged after researchers showed first-generation clocks were too noisy for longitudinal use and rebuilt them as principal-component versions to fix it⁶. Second — and this is the single most important data point in the whole field — DunedinPACE was moved by an actual randomized trial: in CALERIE, two years of sustained caloric restriction slowed DunedinPACE relative to controls, the cleanest randomized evidence that a clock of this type responds to a real intervention⁷. That's a genuine feather in the cap of the clock. It is still not proof about you (more on that below).

So which clock should you pick?

It depends on the question:

• •"How long am I likely to live, relative to peers?" → GrimAge (or GrimAge2) has the strongest mortality-prediction record²³.
• •"How much disease/dysfunction is in my body right now?" → PhenoAge, anchored to clinical-chemistry biology⁴.
• •"Is my rate of aging changing over time?" → DunedinPACE, the only one designed as a rate and the only one a randomized trial has actually shifted⁵⁷.

But notice what every one of those questions has in common: they're about ranking you against a population, or describing a state. None of them is "if I push this number down, will I live longer?" — and that's the question the clocks can't yet answer.

The limit that applies to all three

This is where an honest comparison has to slow the marketing down. The clocks are validated as population-level predictors: the landmark blood-methylation analysis showed people whose methylation runs "old" die sooner on average, even after adjusting for known risk factors⁸. But "associates with mortality across thousands of people" is a categorically different claim from "this is a modifiable target I can move to change my fate." That second claim has not been demonstrated for any clock. An intervention could lower your reading while leaving your real disease risk untouched, and the report would look identical either way.

Two more practical limits stack on top:

• •Test–retest noise. The same sample run twice can return different estimates, and a detailed reliability study found many widely used clocks reproduce poorly — first-generation clocks worst of all⁹. The noise floor on a single re-test can rival the small change a 90-day experiment might produce, so an apparent "improvement" is often measurement variance dressed up as progress.
• •No standard, no regulatory floor. These are laboratory-developed tests, not FDA-cleared diagnostics, and there's no canonical definition of "biological age." Run two reputable clocks on the same person and they can disagree, because they use different training targets, arrays, and reference populations.

The field's own consensus work is explicit about this: it treats these clocks as legitimate but research-stage tools for intervention studies, not as personal diagnostics or treatment-guiding endpoints¹⁰¹¹.

The grades, and how we got there

We grade each clock on the job it was actually built for — and grade the use case most buyers want separately, because conflating them is the marketing trap.

• •GrimAge / GrimAge2 as a population mortality predictor: the strongest of the group — but a predictor, not a proven dial.
• •DunedinPACE as a population pace-of-aging measure: the best design for tracking a rate, and the only clock a randomized trial has moved.
• •PhenoAge as a snapshot of current dysfunction: solid, and uniquely tied to ordinary blood biology.
• •Any clock as a validated guide to YOUR treatment decisions: not there. No clock has been shown to be a target whose movement changes individual outcomes.

That last line is the one that matters for spending decisions. The clocks are real science doing real population work; the "track your protocol quarterly to prove it's working" promise is unearned. We apply exactly this split to the best-known consumer product in our TruDiagnostic TruAge review, and we sort which markers actually earn their place on a panel in what longevity biomarker panels actually test.

Bottom line

There's no universal "best epigenetic clock" — GrimAge wins on mortality prediction, PhenoAge on current dysfunction, and DunedinPACE on tracking a rate of aging (and it's the only one a randomized trial has actually moved). Pick the clock that matches your question. But hold all three to the same honest standard: they rank populations, they're noisy at the individual level, they aren't standardized or FDA-cleared, and not one of them has been validated as a target you can move to extend your own life. Treat any clock number as a soft risk signal, not a scoreboard — and don't reorganize your spending around watching it wobble. For an independently graded look at the labs and clinics selling biological-age testing, see our longevity clinic rankings.