How accurate is a DEXA scan? The actual precision numbers
By The DexaExplained Team · Last verified
Educational body-composition information, not medical advice. Nothing here diagnoses, treats, or prevents any condition, and this site does not interpret bone-density results. Talk to your clinician about anything on a scan report that concerns you.
Repeat a DEXA scan on the same machine the same day and the fat-mass number comes back within about 0.7% of itself. That's the published precision, and it's why DEXA is the reference device in body-composition research. The catch is everything that sentence held constant: the machine, the software, your stomach contents, and your hydration. Change those, and the number moves for reasons that have nothing to do with your body composition.
Precision vs accuracy — different questions
Precision asks: scan me twice, do you get the same answer? Accuracy asks: is that answer the truth? DEXA's precision is excellent and well documented. Its accuracy is judged against the 4-compartment research model, where it ranks at the top of bookable methods — the full ranking is in our method comparison. For tracking your own progress, precision is the number that matters, so that's where the data below focuses.
The published precision numbers
Dordevic et al. (2018) scanned weight-stable adults twice on the same day, then again across three months, on a GE iDXA:
| Measure | Same-day CV | Three-month CV |
|---|---|---|
| Total fat mass | 0.74% | 2.5% |
| Total lean mass | 0.48% | 1.2% |
| Body-fat percentage | 0.80% | 2.5% |
| Visceral fat (VAT) mass | 16.2% | 42.2% |
Two things to take from this table. First, the whole-body numbers are tight — machine noise is under 1%, and even three months of ordinary living only moves a weight-stable adult's readings by a couple of percent. Second, the VAT row is a different sport. A 16% coefficient of variation on same-day scans means a "visceral fat dropped 12%!" result on your report can be pure measurement noise. Judge VAT numbers by trend across several scans, never by one delta.
What moves your number spuriously
Food and water. DEXA can't tell ingested water from lean tissue. In the standardization work by Nana et al., a light meal barely registered — but a meal plus a litre of water shifted apparent fat-free mass by roughly 1.2 kg. Their recommended protocol for research scans: rested, overnight-fasted, minimal clothing, standardized positioning. That's also the cheapest accuracy upgrade you can give your own scan — the prep guide covers it.
Recent exercise. The same standardization work flags daily diet and exercise as activities that move the outputs — the fat-free and lean-mass numbers especially, which is why the protocol specifies rested subjects. Scan before the workout, not after.
A different machine. Vendrami et al. (2024) scanned the same people on a GE Lunar iDXA and a Hologic Horizon A. The Lunar read body-fat percentage about 1.1 points higher and total lean mass about 1.3 kg lower; trunk lean differed by around 12%. The two brands are internally consistent but not interchangeable — the study exists precisely to build conversion equations between them.
Different software. Nana et al.'s reporting standard requires publishing the machine and software version alongside any scan data — both are treated as variables that matter. Practical translation: if your provider upgraded their analysis software between your scans, note it before comparing numbers.
Why DEXA is still the reference
The national body-composition reference data for the United States — NHANES, scanned 1999–2004 across 15 counties — was collected on DXA, and Kelly et al. (2009) turned it into the reference values for ages 8 to 85 that percentile comparisons are built on. When a new device wants credibility, it gets compared to DEXA. No other bookable method has that position.
It isn't perfect. The scan is a two-dimensional projection of a three-dimensional body, water masquerades as lean mass in the ways described above, and the manufacturer gap is real. But its errors are small, documented, and — crucially — stable, which is exactly what a reference method needs.
How to retest so a change is real
- Same machine, same provider. The 1.1-point brand gap is bigger than most 8-week body changes.
- Same conditions: morning, fasted, no workout beforehand, normal hydration.
- Mind the noise floor. Whole-body changes smaller than ~1% of fat mass are within same-day machine noise; VAT changes need to be large or repeated to mean anything.
- Space scans sensibly — how often to scan has the schedule logic.
Already scanned? Paste your report numbers and see what each one means — and which changes clear the noise floor.
Decode your DEXA results →Frequently asked questions
How accurate is a DEXA scan for body fat?
For repeatability — the thing that matters for tracking — published same-day precision in weight-stable adults is a coefficient of variation of 0.74% for total fat mass, 0.48% for lean mass, and 0.80% for body-fat percentage (Dordevic et al. 2018, GE iDXA). Over three months, normal biological variation roughly triples that. The visceral-fat estimate is far noisier: 16.2% CV even on same-day repeat scans.
Why did my body fat change between two DEXA scans?
Possibly because your body changed — but check the conditions first. In standardization testing, eating a meal and drinking a litre of water shifted apparent fat-free mass by about 1.2 kg, because ingested water reads as lean tissue. A different machine brand matters too: scanning the same people on a GE Lunar and a Hologic system produced body-fat percentages about 1.1 points apart. Same machine, same time of day, fasted — then a change is probably real.
Are all DEXA machines equally accurate?
They're all precise, but they don't agree with each other. A 2024 standardization study scanning the same people on a GE Lunar iDXA and a Hologic Horizon A found the Lunar read about 1.1 percentage points higher body fat and about 1.3 kg lower lean mass. Neither is wrong in a useful sense — but numbers from different manufacturers aren't interchangeable, so never judge progress across brands.