Epigenetic Age Tests Compared: Horvath, GrimAge, PhenoAge Explained
Three major epigenetic clocks compared — what each measures, accuracy, cost, which to choose, and the test-retest problem that no provider talks about openly.
Steve Horvath was a UCLA biostatistician in 2013 when he published one of the most cited papers in aging biology. He showed that methylation at 353 specific sites on the genome could estimate biological age across most human tissues with remarkable accuracy. The first epigenetic clock. The longevity community treated it as a breakthrough — for the first time, biological age was a measurable number, not a vibe.
A decade later, three clocks dominate the commercial market, and the consumer testing industry built around them generates real revenue selling $300 saliva tests to men trying to find out how old they really are. The technology is real. The interpretation is more complicated than the marketing suggests.
What Epigenetic Clocks Actually Measure
DNA does not change during a lifetime. The chemical modifications layered on top of it do. Methylation is the addition of a methyl group (–CH₃) to specific cytosine bases on the DNA, typically at CpG sites. Methylation patterns shift predictably across the genome as cells age — some sites gain methylation, others lose it, and the cumulative pattern correlates tightly with chronological age and, more importantly, with biological aging.
An epigenetic clock is a mathematical model. Researchers measure methylation at hundreds to thousands of CpG sites, then train a regression model that takes those methylation values as inputs and outputs an age estimate. Different clocks use different CpG sites and different training data — which is why a single sample run through multiple clocks produces different numbers.
The major commercial clocks split into two generations:
First generation — trained against chronological age. Horvath 2013 is the prototype. Accurate at estimating birth-certificate age in healthy populations but limited at detecting departures from healthy aging.
Second generation — trained against clinical biomarkers and mortality outcomes. PhenoAge (Levine 2018) and GrimAge (Lu 2019) fall here. These clocks are better at predicting time-to-death and disease onset, which is what most people actually want from a "biological age" measurement.
Third generation (rate of aging) — DunedinPACE (Belsky 2022) and similar. Rather than estimating cumulative age, these clocks estimate the current pace of aging — how fast biological aging is occurring at the moment of measurement. Useful for tracking intervention response over months rather than decades.
Horvath 2013: The Original
Horvath's clock uses methylation at 353 CpG sites and produces an estimate calibrated to chronological age across nearly all human tissue types (blood, saliva, organ biopsies). Correlation with chronological age in the original training data was 0.96.
What it's good for: tissue-agnostic biological age estimation in research, comparing age across different tissues from the same individual, historical baseline since most published longevity intervention studies use Horvath.
What it's not good for: predicting time-to-death or detecting subtle responses to lifestyle interventions. Because Horvath was trained against chronological age, deviations from chronological age have weaker correlations with mortality and disease outcomes than the second-generation clocks.
Consumer availability: included in most multi-clock reports from TruDiagnostic, MyDNAge, and others. Rarely sold alone.
Levine PhenoAge 2018: The Biomarker Correlate
Morgan Levine and colleagues at Yale built PhenoAge by first developing a "phenotypic age" composite from nine standard clinical biomarkers (albumin, creatinine, glucose, CRP, lymphocyte percent, mean cell volume, red cell distribution width, alkaline phosphatase, white blood cell count), then training a methylation clock against that composite.
The result: a clock that correlates with clinical aging markers more tightly than Horvath. PhenoAge predicts all-cause mortality, cardiovascular events, and cancer incidence beyond what chronological age alone explains. People with elevated PhenoAge relative to chronological age have substantially elevated event rates.
What it's good for: tracking systemic biological age in the context of clinical biomarkers. PhenoAge captures inflammatory and metabolic aging well. Useful for men interested in healthspan rather than just lifespan.
What it's not good for: maximum mortality prediction precision (GrimAge does better).
Lu GrimAge 2019: The Mortality Predictor
Ake Lu and Steve Horvath developed GrimAge by training a methylation clock against time-to-death using DNA methylation surrogates for seven plasma proteins (including PAI-1, GDF15, leptin) and self-reported pack-years of smoking. The result is the most accurate epigenetic predictor of all-cause mortality and time to disease onset currently available.
GrimAge predicts time to coronary heart disease, cancer, type 2 diabetes, and cognitive decline beyond what chronological age, PhenoAge, or Horvath provide. In the original validation, GrimAge acceleration was strongly associated with mortality even after adjustment for chronological age, sex, and major risk factors.
What it's good for: longevity-focused interpretation. If the question is "how much risk of premature death do I currently carry," GrimAge is the clock that answers it most directly.
What it's not good for: tracking intervention response over weeks to months (DunedinPACE is more responsive). GrimAge moves slowly because it integrates a lifetime of biological history.
Consumer availability: GrimAge is licensed and available through TruDiagnostic and some research-grade providers. Not all consumer epigenetic tests include it.
Epigenetic clocks are research tools sold as consumer products. The numbers they produce are real signals embedded in real noise. Treat them as one input among several, not as a verdict.
DunedinPACE: The Rate-of-Aging Clock
A newer addition based on the longitudinal Dunedin Study cohort, DunedinPACE estimates the current pace of biological aging in years-of-aging per chronological year. A score of 1.0 means you're aging at one biological year per chronological year. A score of 0.8 means you're aging 20% slower than normal; 1.2 means 20% faster.
DunedinPACE is the most responsive to short-term interventions because it measures the current rate rather than cumulative damage. Bryan Johnson uses DunedinPACE as one of his primary tracking metrics — and reports a sustained pace around 0.65–0.69, meaning his current aging rate is about 30% below normal.
The Test-Retest Problem
The field's underdiscussed weakness: precision. Higgins-Chen et al. (2022) ran multiple methylation samples from the same individuals and demonstrated test-retest standard deviations of 2–5 years for most published clocks. Same person, same blood draw split into two tubes, can produce different epigenetic age estimates 2–5 years apart.
This matters operationally. If your "biological age" drops 3 years between this year's test and next year's test, you cannot confidently attribute that to interventions. It might be noise. The signal is real but the variance is wider than the marketing implies.
Higgins-Chen et al. also developed "principal component" versions of the clocks (PCHorvath, PCPhenoAge, PCGrimAge) that reduce technical variance substantially. Some commercial providers now report PC-adjusted versions; ask whether yours does.
The Consumer Test Landscape
The major commercial providers as of 2026:
TruDiagnostic (TruAge) — multi-clock reporting including PCHorvath, PCPhenoAge, PCGrimAge, and DunedinPACE. Around $499 for the Complete panel. Generally considered the most rigorous consumer option.
MyDNAge (Zymo Research) — based on the Horvath skin and blood clock. Around $299. Simpler reporting, longer trajectory of consumer availability.
Elysium Health (Index) — multi-clock test based on a proprietary methylation panel. Around $299. Less transparency on exact clock methodology than TruDiagnostic.
Hannum Clock providers — some labs still offer the Hannum 2013 blood-specific clock, which has good blood-tissue accuracy but limited use over the newer second-generation clocks.
Sample type matters. Most consumer tests use a small saliva or finger-stick blood sample. Blood samples generally produce more reliable methylation data than saliva, though modern saliva protocols are acceptable for most clocks. Whole blood reflects immune cell aging more directly than saliva.
Who Should Bother and Who Shouldn't
Epigenetic testing makes sense for:
- Men over 40 establishing a longevity baseline alongside comprehensive bloodwork and DEXA
- Anyone running structured longevity interventions who wants a high-level integrative marker
- Researchers, self-experimenters, or N-of-1 tracking enthusiasts
Epigenetic testing is lower priority for:
- Men with no measurable health complaints whose comprehensive bloodwork and fitness markers are already optimized — the epigenetic test mostly confirms what better measurements already told you
- Anyone running tight budgets — VO2 max test, DEXA, and quarterly bloodwork produce more actionable data per dollar
- People who will use one elevated reading as the basis for unproven interventions — the noise is too high for that
The integration matters. A single epigenetic age reading in isolation tells you less than the same reading paired with VO2 max, ApoB, fasting insulin, lean mass index, and HRV. The clocks are integrative markers; use them as integrative inputs.
The Protocol
- Start with the foundation, not the test — VO2 max test, DEXA scan, comprehensive bloodwork (ApoB, fasting insulin, hs-CRP, full panel). Epigenetic testing comes after, not before.
- Choose a multi-clock provider — TruDiagnostic's TruAge Complete is the most comprehensive consumer option in 2026 ($499); it reports PC-adjusted Horvath, PhenoAge, GrimAge, and DunedinPACE. MyDNAge or Elysium at $299 are reasonable lower-cost alternatives.
- Standardize sample conditions — same time of day, similar fasting and hydration state if doing repeat tests, same provider for comparability.
- Pay attention to PC-adjusted scores — these have lower technical variance than the original clock versions. Ask the provider which version they report.
- Track DunedinPACE for short-term intervention response — if you're running a 12-month protocol and want to know if it's working, DunedinPACE is more responsive than the cumulative clocks.
- Retest at 12 months minimum — more frequent testing produces noisier data without meaningful signal. Treat changes under 2 years as ambiguous.
- Never make a major intervention decision on one epigenetic reading alone — triangulate with clinical biomarkers, functional tests, and subjective indicators. The clocks are signals embedded in noise.
Want to combine your epigenetic age results with comprehensive lifestyle and biomarker inputs? → Use the PrimalPrime Biological Age Calculator for an integrated longevity score and intervention sequence.