How Accurate Is a DEXA Scan? (What the Number Actually Means)

TL;DR: DEXA is the best practical body composition tool most people can buy, and it still isn't truth. It's highly repeatable, which is what you actually need for tracking, but the absolute number carries several percentage points of uncertainty and doesn't transfer between machines. Scan under identical conditions, use the same scanner every time, and ignore small moves.

---

Someone pays $80, lies on a table for seven minutes, and walks out with a number to one decimal place. 18.4% body fat. It feels like a lab result. The decimal point is doing a lot of persuasive work there, and it shouldn't.

DEXA (dual-energy X-ray absorptiometry, often written DXA) is genuinely the most useful body composition measurement available at consumer prices. But "most useful" and "correct" are two different claims, and the gap between them changes how you should use your results.

Precision and accuracy are not the same question

Precision is: if I scan twice, do I get the same answer? Accuracy is: is that answer right?

DEXA does very well on the first and merely okay on the second.

For same-day repeat scans where the subject gets off the table and gets back on, published precision error for total body fat percentage generally comes in around 1 to 2 percent, with total fat mass coefficients of variation in the 1 to 3 percent range. Regional measurements (arms, legs, trunk separately) are noticeably worse than whole-body numbers (Nana et al., 2015).

That's a good result. It means DEXA can detect a real change in your body composition far better than a scale, a caliper, or a smart scale can. Our rundown of body fat measurement methods puts the alternatives side by side, and DEXA wins the repeatability comparison easily.

Accuracy is a different story. When DEXA gets compared against a four-compartment model (the research criterion, which measures body water and bone mineral separately instead of assuming them), individual agreement is much looser than the decimal point suggests. Limits of agreement in the range of plus or minus 4 to 5 percentage points of body fat are common in validation work, and the direction of the error is not random: some systems systematically underestimate fat mass relative to criterion methods (Van Der Ploeg et al., 2003; Schoeller et al., 2005).

So your 18.4% might be 16%. It might be 22%. What it almost certainly is, though, is 18.4% on that machine, under those conditions, which is the part you can use.

Why the absolute number drifts

Three things push DEXA around, and none of them are obvious from the printout.

It assumes lean tissue hydration is constant. DEXA can only separate two things at a time from an X-ray attenuation signal, so it treats the non-bone soft tissue as a fixed mix of water, protein, and mineral. Lean tissue is assumed to be roughly 73% water. When you're dehydrated, carb-depleted, or glycogen-loaded, that assumption breaks, and the software reports the change partly as a change in lean mass. This is the same mechanism that makes the bathroom scale lie to you day to day, covered in why the scale lies about water weight.

Machines disagree with each other. A Hologic and a GE Lunar scanner will not give you the same body fat percentage on the same person on the same day, and cross-calibration equations exist specifically because of this. Software version upgrades on the same hardware can also shift results (Shepherd et al., 2012). A "DEXA scan" is not a standardized unit the way a kilogram is.

Tissue depth matters. X-ray attenuation depends on how much tissue the beam passes through, so very lean people, very large people, and people whose body thickness sits outside the calibration range get less reliable estimates. The error isn't uniform across the population.

What you do before the scan changes the result

This is the part almost nobody controls, and it's the biggest source of avoidable noise.

Nana and colleagues ran a series of studies on exactly this in active people, scanning subjects under different real-world conditions rather than standardized ones. Eating a meal, drinking fluid, or training before the scan all shifted the numbers. A meal alone was enough to move estimated lean mass by around a kilogram, because the food and fluid in your gut is soft tissue as far as the machine is concerned, and it gets counted as lean (Nana et al., 2012).

A kilogram of phantom lean mass is more than most people gain in two months of hard training. If you ate breakfast before one scan and fasted before the next, your "progress" is partly breakfast.

The protocol that the research group settled on is the one you should copy:

• Overnight fasted, first thing in the morning
• No fluid before the scan
• No training in the 24 hours prior, and definitely not that morning
• Bladder emptied immediately before
• Minimal clothing, no metal
• Same machine, same operator, same positioning every single time

That last point is doing most of the work. If you hold conditions constant, you convert DEXA from a mediocre absolute measurement into a very good relative one.

How much change counts as real

Here's the practical rule that follows from the precision data. If a single scan carries roughly 1 to 1.5 percent body fat of measurement error, then the difference between two scans carries more than that, because both scans have error.

In practice: treat anything under about 1.5 percentage points of body fat, or under roughly 1 kg of fat mass, as indistinguishable from noise. A move from 18.4% to 17.6% over six weeks is not evidence of anything. A move from 18.4% to 15.1% over four months is real.

This is why scanning monthly is mostly a waste of money. The signal you're looking for takes a quarter to accumulate past the error bars. Scan two to four times a year, at the bookends of a cut or a bulk, and use cheaper daily and weekly signals in between. How to track recomp progress covers the in-between layer: scale weight trend, waist and limb measurements, progress photos, and strength at a fixed bodyweight, which together tell you more than a quarterly scan does on its own.

Where DEXA still beats everything else

None of this is an argument against getting one. Compared to the alternatives, DEXA is still the best consumer option, and it's not close.

Bioimpedance devices, including the InBody machines at most gyms, use the same hydration assumption as DEXA but with a far weaker signal, which is why they swing more between sessions. We went through that in detail in how accurate an InBody scan is. Consumer smart scales are worse again, as covered in are smart scales accurate for body fat. Calipers are operator-dependent to the point that two technicians can differ by several percent on the same person.

DEXA also gives you two things the others don't: a regional breakdown that shows where fat and lean mass actually sit, and a bone mineral density reading that has clinical value on its own, especially if you've been dieting hard or training for a long time.

If you want a number today without the appointment, a body fat calculator built on tape measurements will get you within a few percent, which is honestly the same neighborhood of uncertainty as a single unstandardized DEXA. It just won't track change as cleanly.

The honest way to read your result

Treat the body fat percentage as an estimate with a range attached, not a fact. Treat the change between two well-standardized scans on the same machine as real information. Never compare your DEXA number to a friend's from a different clinic and conclude anything.

And remember what the number is for. Body fat percentage isn't the goal. It's a proxy for whether the fat you're losing is fat and the lean mass you're building is lean mass, which is the thing that actually determines how a cut or a bulk is going. Lean body mass is the underlying variable, and it's the one worth defending.

A scan twice a year tells you whether the last five months worked. It can't tell you whether this week is working. For that you need the daily inputs (weight trend, intake, training load) feeding something that models the trajectory instead of just recording it. That's exactly what Protokl does between scans: it takes your logged weight, food, and training, separates signal from water noise, and projects where your body composition is actually heading, so your next DEXA confirms what you already knew instead of surprising you.