Guide · 6 min read
How accurate are body fat estimates?
Every body fat number is an estimate. The useful question is not “which method is exact” — none is — but how far off it is, and whether it can detect change in you.
Updated · 4 sources · Not medical advice
Two different kinds of error
Accuracy is how close a method gets to a reference measurement, on average and for one person. Repeatability is how close it gets to itself when nothing changed. A method can be several points off for you and still be good at showing that you changed — if it is off by the same amount each time.
For tracking progress, repeatability matters more. For a one-time snapshot, accuracy does.
Method by method
| Method | Typical error vs. DXA | Good for tracking change? | Watch out for |
|---|---|---|---|
| DXA scan | It is the usual research reference | Yes, on the same machine | Different machines and software give different numbers; hydration and recent meals shift it |
| Smart scale (bioimpedance) | Varies a lot by device | Only under identical conditions | Water, food, exercise and time of day move the reading |
| Tape formulas (US Navy, RFM) | About 3–4 points on average; more for some people [1][2] | Poor for small changes over weeks [3] | Measuring site and tape tension |
| Validated phone scan | About 3.5 points on average [4] | Better: repeat error about 0.5–0.7 points [4] | Only the specific validated app, with its capture rules |
| Single photo judged by AI | No published validation against DXA we could find | No: the same photo can get different answers | Lighting, pose and clothing |
DXA is a reference, not the truth
DXA is the most common reference in studies, which is why everything else is compared with it. It is still an estimate: results depend on the scanner model, software version and your hydration. Repeat DXA scans on the same machine are very consistent, which is what makes DXA useful for research.
What about photo apps?
The best-studied smartphone scan we know of, tested against DXA in 131 adults, averaged about 3.5 points of error — similar to a tape formula — but repeat scans differed by only about half a point [4]. That means it can show change of roughly 1.5–2 points with some confidence, while a single number still carries a wide range.
General-purpose AI that “reads” body fat from one photo is a different thing. We found no peer-reviewed validation against DXA, and language models can give different answers to the same image. A number that changes when nothing changed can’t track progress.
Tensile does not estimate body fat from photos. In this version the photo never leaves your phone; your tape measurements are the anchor, and every number shows its True Range. See How we measure.
How to read your number
- Always read it as a range. “About 22%, likely 18–26%” is the honest version of “22%.”
- Compare a method only with itself, under the same conditions.
- Count a change only when it is bigger than the method’s noise.
- Track your waist and strength alongside it. They are cheaper, more repeatable and directly trainable.
Questions
- What is the most accurate way to measure body fat?
- In practice, a DXA scan or a multi-compartment lab method. Even those have error, and results vary between machines. For tracking at home, a consistent method used the same way matters more than the method itself.
- Why does my smart scale say something different every day?
- Bioimpedance scales are sensitive to hydration, food, exercise and skin temperature. Weigh-ins under the same conditions help, but day-to-day swings are mostly water, not fat.
- Can an app tell my body fat from a photo?
- Only a few scan apps have published validation against DXA, and even those average a few points of error. A single photo read by a general AI model has no such validation and can give different answers to the same picture.
Sources
- 1.Hodgdon JA, Beckett MB. Prediction of percent body fat for U.S. Navy men and women from body circumferences and height. Naval Health Research Center, Reports No. 84-11 and 84-29; 1984.
- 2.Woolcott OO, Bergman RN. Relative fat mass (RFM) as a new estimator of whole-body fat percentage — a cross-sectional study in American adult individuals. Sci Rep. 2018;8:10980. doi.org/10.1038/s41598-018-29362-1
- 3.[Circumference-based body-fat equations compared with reference methods for tracking change over a short training program.] Front Physiol. 2023. doi.org/10.3389/fphys.2023.1183836
- 4.Tinsley GM, et al. [Validation of a smartphone body-composition scan against DXA in 131 adults: mean absolute error about 3.5 percentage points; repeat-scan error about 0.5–0.7 points.] Front Med. 2024. doi.org/10.3389/fmed.2024.1485450