If you take two different biological age tests, you will likely get two different results. Different clock targets are one reason — but measurement error, calibration differences, sample handling, algorithm versions, and how well the population you belong to fits the clock's training data can all contribute too. Neither test is necessarily “wrong.”
This is the most common source of confusion in the biological age space — and the most important thing to understand before you compare products, interpret results, or track changes over time.
Nearly all aging clocks fall into one of three paradigms. Each is trained on a different target and has a different intended use.
Paradigm 1: How old does your body look?
The first generation of aging clocks were trained to answer a straightforward question: given your biological data, how old do you appear?
These are sometimes called chronAge clocks or age-prediction clocks. The most well-known are Steve Horvath's pan-tissue clock (2013) and Gregory Hannum's blood-based clock (2013). Both were trained by feeding DNA methylation data from thousands of people into a model alongside their known calendar ages. The model learns which methylation patterns correspond to which ages — and then predicts age for new samples.
When your predicted age deviates from your actual age, researchers call that age acceleration (or deceleration) — an estimate relative to the model and its reference data at that single point in time. Someone who is 50 but has a predicted biological age of 55 has an elevated age-acceleration estimate; that is not the same as a measured rate of aging, which would require repeated measurements over time.
The strength of these clocks is precision in age estimation — the Horvath clock predicts chronological age with a median error of about 3.6 years. The limitation is that predicting calendar age well doesn't necessarily mean predicting health outcomes well. A person can look biologically “young” on a chronAge clock while still carrying significant disease risk.
Paradigm 2: What is your health risk?
Second-generation clocks took a different approach. Instead of training on calendar age, they were trained on health outcomes — mortality, disease incidence, or composite health measures.
PhenoAge (Levine, 2018) was trained on a combination of clinical biomarkers known to predict mortality, then mapped back to DNA methylation. GrimAge (Lu, 2019) is built from DNA-methylation surrogates for smoking pack-years and for a set of plasma proteins, rather than measuring smoking history or the proteins themselves directly, and has performed well as a population-level predictor of lifespan and healthspan in the cohorts where it has been validated.
These clocks sacrifice some accuracy in age prediction to gain predictive power for what most people actually care about: health and longevity. A person who looks “young” on the Horvath clock but “old” on GrimAge may have chronAge-typical methylation patterns but elevated risk factors that the first-generation clock wasn't designed to detect.
Paradigm 3: How fast are you aging right now?
The newest paradigm doesn't estimate a cumulative biological age at all. Instead, it measures the current rate of aging — how many years of biological wear your body accumulates per calendar year.
DunedinPACE (2022) is a well-known example of this approach. It was developed from the Dunedin longitudinal study, which tracked the same group of people from birth into middle age, measuring 19 biomarkers repeatedly over two decades. The resulting clock captures the pace of biological change rather than an accumulated state.
A pace of 1.0 represents the norm within DunedinPACE's own reference and training framework, not a universal population average that applies outside it. Below 1.0 means slower than that reference; above 1.0 means faster. A 2023 post hoc analysis of the CALERIE randomized trial found a small change in DunedinPACE among participants in the caloric-restriction arm, but not in PhenoAge or GrimAge measured in the same trial; the analysis did not establish a clinical benefit. That a score moves in one study is evidence the clock can register something in that context — it is not proof that a given intervention improved health or function.
Pace-of-aging clocks answer a different question than age-prediction or health-risk clocks: they describe a current trajectory rather than an accumulated state. A person who aged rapidly for a decade but recently slowed down could show a favorable pace at a single point in time — the two paradigms simply aren't measuring the same thing.
Why one person gets different results
This is now easy to understand. A 50-year-old might receive:
- Horvath clock: biological age 47 — their methylation patterns look younger than average.
- GrimAge: biological age 54 — their mortality-linked markers suggest elevated risk.
- DunedinPACE: 0.92 — they are currently aging slower than average.
These results don't necessarily contradict each other — each clock is reading a different dimension of the same biology, and all three can be simultaneously true. The Horvath result says their methylation landscape looks young. GrimAge says their health risk profile looks older. DunedinPACE says their current trajectory is favorable. But disagreement between clocks isn't always explained by different targets alone. Measurement error, calibration differences, the sample or algorithm version used, and differences between the population a clock was trained on and the person taking the test can all contribute to two clocks giving results that look inconsistent.
Which paradigm matters for what
The three paradigms serve different purposes, and matching the right clock type to your goal matters more than picking the one with the most impressive marketing.
Age-prediction clocks are useful as a general benchmark — a reference point for where you stand relative to the population. They are well-validated and widely used in research, but they weren't designed to predict disease or respond sensitively to lifestyle changes.
Health-risk clocks carry more population-level prognostic weight than age-prediction clocks — they were trained directly on mortality and disease outcomes, so a GrimAge or PhenoAge result says more about risk trends observed across similar cohorts. That is not the same as individual diagnostic or clinical utility: a result tells you how your markers compare to population patterns, not a clinical diagnosis, and it is not a substitute for evaluation by a clinician.
Pace-of-aging clocks are oriented toward tracking change over time rather than a one-off snapshot. A 2023 post hoc analysis of the CALERIE randomized trial found a small change in DunedinPACE, though not in PhenoAge or GrimAge, in the same trial — a research signal that the clock can register something, not evidence that any given intervention improved health or function.
What to ask before you test
Before choosing a biological age test, the most useful question isn't “which test is best?” It's “what do I want to know?”
If you want a baseline snapshot, a chronAge clock's target may match your question. If you want a population-level health risk signal, a GrimAge or PhenoAge target may fit better. If you want to track whether a change you've made is registering biologically, DunedinPACE's target is oriented that way — though a favorable reading is a research signal, not proof of a health benefit. Matching the target to your question is only the first step: how any specific consumer test implements that clock, its repeatability, its calibration, and what evidence the provider publishes all still determine whether a given result is actionable. And if you want the most complete picture, some providers now offer panels that include clocks from more than one paradigm.
Understanding the paradigm behind a clock is the difference between being confused by conflicting results and having a reasonable sense of what each number is, and isn't, telling you. See our deeper look at reading a result without overreading it for how to interpret any single result responsibly.
Sources
These are the primary papers behind the clocks discussed above. They establish the clocks' validation as research instruments — they do not by themselves establish clinical utility for an individual consumer test result.
- Horvath (2013): DNA methylation age of human tissues and cell types
- Hannum et al. (2013): Genome-wide methylation profiles reveal quantitative views of human aging rates
- Levine et al. (2018): An epigenetic biomarker of aging for lifespan and healthspan (PhenoAge)
- Lu et al. (2019): DNA methylation GrimAge strongly predicts lifespan and healthspan
- Belsky et al. (2022): DunedinPACE, a DNA methylation biomarker of the pace of aging
- CALERIE trial post hoc analysis (2023): a small, exploratory effect on pace-of-aging clocks, not a validated clinical benefit