What a smart scale's body-fat number is worth against the reference scan

A smart scale or a smartwatch that reports your body-fat percentage does it by bioelectrical impedance — passing a tiny current through the body and inferring composition from how that current is resisted. Against the reference standard, a dual-energy X-ray absorptiometry (DXA) scan, a recent validation study found a wrist device tracked body-fat percentage with roughly 14% average error and did so more accurately than a clinical impedance analyser — but its estimate of muscle mass was much weaker, and a companion study showed the reading moves depending on whether you are sitting, standing or lying down [s1][s2].

The takeaway is not that these numbers are useless, but that they answer one question better than another, and that consistency of technique matters as much as the device.

Body fat: usable; muscle mass: not really

A study in Frontiers in Sports and Active Living measured 108 physically active adults (56 women, 52 men) with a wrist bioimpedance device (Samsung Galaxy Watch5), a standing clinical analyser (InBody 770), and DXA as the criterion [s1]. For body-fat percentage, the wrist device correlated strongly with DXA (r = 0.93; concordance CCC = 0.91) and had a mean absolute percentage error of 14.3%, compared with 21.1% for the clinical analyser [s1]. In women specifically, the wrist device was more accurate still, with a body-fat error of 9.19% and statistical equivalence to DXA supported [s1].

Muscle mass was a different story. Although correlations looked strong, agreement was classified as weak — concordance of just 0.45 for the wrist device and 0.25 for the clinical analyser, with errors of 20.3% and 36.1% respectively [s1]. In other words, the device that gives a serviceable body-fat figure gives a muscle-mass figure that should not be trusted at the individual level, which matters for anyone tracking muscle loss in ageing or trying to preserve muscle during weight loss, where the muscle number is the whole point.

The study also flagged proportional bias: the error grew in people with higher body fat, so the reading drifts furthest from the truth exactly where the stakes of an accurate number are often highest [s1].

Posture alone changes the answer

Impedance depends on how fluid is distributed in the body, and that distribution shifts with posture. A separate study of 117 adults found that smartwatch bioimpedance estimates of body fat, fat mass and fat-free mass differed significantly between the standing and supine positions — with the effect concentrated in women — even though group-level agreement with DXA stayed broadly similar across positions [s2]. The authors' conclusion was practical: standardise body position when using these devices, because an unstandardised technique introduces error on top of the device's own [s2].

That is the difference between a number you can compare with last month's and a number you cannot. Measured the same way each time — same posture, similar hydration, similar time of day — a wrist or scale reading can track a trend even if its absolute value is off. Measured casually, the day-to-day swing can be an artefact of how you stood on the scale.

Why body fat reads better than muscle

The split in the results is not random. Bioimpedance estimates fat indirectly: fat is a poor conductor, so a body with more fat resists the current more, and the device maps that resistance onto a body-fat figure through population equations. Body fat is a large, relatively stable share of the body, so the mapping is forgiving — which is why both the wrist and clinical devices correlated strongly with DXA on body-fat percentage and why the wrist device's error, at 14.3%, beat the clinical analyser's 21.1% [s1]. Skeletal-muscle mass is a smaller quantity that the device must partly derive by subtraction, so small errors in the impedance reading translate into large errors in the muscle figure — the concordance of 0.45 and 0.25 the study reported reflects that fragility [s1]. It is also worth noting what impedance cannot see at all: it does not distinguish visceral from subcutaneous fat, and its readings move with hydration, so a glass of water or a workout can shift the estimate independently of any real change in composition.

How to read the number

Treat a consumer body-composition reading as a trend tool for body fat, taken under consistent conditions, and as broadly indicative rather than exact — the wrist device's 14% average error means an individual estimate can sit several percentage points from the truth [s1]. Do not rely on its muscle-mass figure to make decisions, given the weak agreement the validation found [s1]. And hold hydration, meals and posture constant between measurements, because the study evidence shows those variables alone can move the result [s2]. For any clinical purpose, the scan, not the scale, remains the measurement.