Are Body Fat Scales Accurate? One Reading Can Be 14.5 Points Off

The scale says 22.4%. It has a decimal place, which is a confident thing for a bathroom scale to have.

So here is the question, split the way the research splits it. Is that number accurate on average, across a lot of people? Yes, reasonably. Is it accurate for you, this morning? That is a different question, and the answer is a good deal wider than one decimal place.

What the scale is actually doing

It is not measuring fat. It is measuring water.

A body-fat scale sends a small, harmless electrical current up one leg and down the other. Water conducts electricity; fat does not, much. So the scale measures how much the current was resisted — the impedance — and from that, together with your height, weight, age and sex, an equation estimates how much of you is water, then how much is lean tissue, then, by subtraction, how much is fat. The method was validated for fat-free mass in the mid-1980s, in a lab, with electrodes on a hand and a foot [13]. A scale you simply stand on uses the feet only; some home scales add a handle you grip, which sends the current through the arms and upper body as well [1].

Every step of that is an estimate stacked on an estimate. Which is fine — the question is only how big the stack gets.

The test: three bathroom scales against an MRI

A 2008 study of ordinary consumer scales put 106 adults aged 22 to 78 — normal weight through obese — on three foot-to-foot bathroom scales and one hand-to-foot scale with a handle to grip, and compared every reading against whole-body MRI and DXA, the two reference methods [1].

On average, the scales were fine. Three of the four sat within 1.5 points of the reference; the correlations were strong [1]. If you lined up a hundred people and wanted the group’s average body fat, a bathroom scale would get you there.

Then look at the individual rows. On the worst of the three foot-to-foot scales, a single reading could sit anywhere from 8.6 points below the DXA value to 14.5 above [1]; on the best of them, from about 6 below to about 7 above. That is the band inside which 19 of 20 readings land. So on that worst scale, a display that says 22% is compatible with a DXA reading of about 7, or about 31. The authors’ own conclusion: foot-to-foot scales are for “group estimation”; for an individual, use the hand-to-foot kind [1]. One thing to weigh beside that advice: the study was paid for by the company that made the hand-to-foot device it tested, and one of the foot-to-foot scales, and one of its authors worked there [1].

A 2023 test went wider: fifteen impedance devices, fourteen of them consumer models, in 73 adults aged 19 to 50, against a four-compartment lab model, the reference that measures water, bone and body density separately [15]. On the ten scales you simply stand on, the band 19 readings in 20 fall inside ran from plus or minus 9.3 to plus or minus 14.7 points around each scale’s own average error; on the three consumer devices you grip as well as stand on, plus or minus 6.6 to 8.8 [15]. On the group average, fourteen of the fifteen were within 3.6 points, but one stand-on scale read 11.7 points high on average, and only five of the fifteen were close enough to count, statistically, as matching the lab model within 2 points [15]. Right on average, then, for most scales but not all of them; wrong for one person on every scale you simply stand on.

That is the whole page in one table. What follows is why.

The error has a direction

If the scale were simply noisy — random, sometimes high, sometimes low — averaging a few readings would fix it. It is not simply noisy.

Which way it leans depends on the device and the person. In the bathroom-scale study, all three foot-to-foot scales read high in people with less body fat and low in people with more; the hand-to-foot device showed no such tilt [1]. A study that tracked 48 active adults across six sessions on a single-frequency analyser found the reverse tilt, reading fat low in leaner people, with a disagreement that was larger in women and depended on body size enough that the author concluded no blanket statement about the two methods could be made [2]. In children at ordinary schools, a manufacturer’s equation ran an average of 5.2 points low; in girls at a sports academy, the same equation ran 1.8 points high [3].

Averaging does not remove a bias. It just gives you the same wrong number more precisely.

A glass of water moves it half a point to a point

Seventy-six young adults, average age 21, were measured on a hand-to-foot analyser, then drank 20 oz (591 ml) of water, then were measured again [5]. Twenty minutes later their body fat read 1.1 points higher — nothing about their bodies had changed except 20 oz of water in the stomach. The authors put the rise down to two things: the analyser weighed the water as extra body mass, and the impedance went up, which the equation reads as less lean tissue. The effect was largest in the lightest people [5]. On a foot-to-foot analyser, the arrangement a bathroom scale uses, the same 20 oz left the impedance unchanged in 21 young men, and the reading still rose about half a point, from the added weight alone, which the authors said may matter little in practice [16].

Push it harder and it gets stranger. Give people about 28 oz of water (11 ml per kg — that is the dose for someone weighing 165 lb) and stand them still for an hour, and a multi-frequency clinical analyser registered the added weight entirely as fat: about 2.9 lb (1.3 kg) of it [6]. Simply standing for the hour drifted the leg readings on its own.

A day of eating does the same in the other direction: after meals, impedance dropped and the fat-mass estimate fell by 1.4 to 2.4 percent of itself on average (a share of the fat figure, not percentage points of body fat), with some people’s estimates falling by as much as a tenth — and unlike DXA, the impedance readings had not settled back by the next morning’s fast [14].

Two studies pull the other way and belong here. One found food and drink shifted a foot-to-foot reading by about a point, which its authors judged “within the imprecision of the impedance technique” and not clinically meaningful [7]. Another found a 400-calorie breakfast made no clinically relevant difference on a hand-to-foot device [8]. Read together with the rest: a point or two of wobble from what you ate and drank is real, it is smaller than the device’s own error, and it is exactly the size of the changes people get excited or despairing about week to week.

Can it at least track my progress?

This is the fallback everyone reaches for. Fine, the number is off — but if it is off consistently, I can watch the trend. Two studies checked: one over 21 weeks in older adults, and one over 12 to 16 weeks on fifteen devices, most of them scales you simply stand on.

Seventy-two adults over 65, some exercising, some not, measured at weeks 1, 12 and 24 by nine different impedance equations and by DXA, all against a four-compartment lab model — the closest thing to ground truth that exists [9]. Single readings correlated well, as they always do. Then the researchers asked the only question that matters for tracking: for each individual, did the device get the change in fat-free mass (everything that isn’t fat) right?

Total agreement on individual change in fat-free mass: 25% for impedance. And 27% for DXA [9].

Sit with the second number. The method most people treat as the gold standard tracked an individual’s change in lean mass correctly about a quarter of the time, in a 21-week study of adults over 65, against the real reference. Impedance was not meaningfully worse, and that was a research analyser run through nine different equations, not a bathroom scale. This is not only a scale problem. It is a measuring-one-person problem.

A test of the scales themselves came in 2023. Thirty-seven of the 73 adults in the fifteen-device study came back 12 to 16 weeks later, and each device’s reading of how their body fat had changed was checked against the four-compartment model’s [15]. On the group’s average change, nine of the fifteen devices matched the lab model, statistically, to within a point. For one person it was another story. On every device, the research-grade one included, the band 19 readings in 20 fall inside ran from plus or minus 3.4 to plus or minus 5.2 points on the change itself, while the real changes in the group ran from 5 points down to 7.5 up and averaged 0.2 [15]. Six of the stand-on scales also squeezed the changes they saw: the people whose body fat moved most were shown a smaller move than the lab model measured [15]. The authors’ reading: some models, mostly the ones you grip as well as stand on and a few stand-on scales, may be useful for tracking, and about half the devices did poorly enough that home users should be wary [15].

What the evidence supports is narrower and still useful. Measured under the same conditions each time, a move bigger than that band, about five points, is unlikely to be the device’s error alone. That is our reading of the band, not something the study tested: the largest real change in it was 7.5 points, and the average 0.2. What the evidence does not support is reading anything into a two-point move over a fortnight: two points sits inside the band on every device tested [15].

What about muscle mass?

Worse. Same physics, more assumptions. A 2025 trial put a wrist-worn BIA watch and a clinical hand-to-foot analyser against DXA in 108 active adults [4]. For body fat, both correlated strongly, with errors of 14% and 21% of the true value. For muscle mass, the correlations were still high but the agreement was rated weak on the standard scale — the devices and DXA tracked each other without landing on the same number [4]. If the scale tells you that you gained a pound of muscle this week, that is a sentence the technology cannot support.

So what is the most accurate way to measure body fat?

In a lab: a four- or five-compartment model that separately measures your water, your bone mineral and your body density and combines them. Nobody has one at home.

DXA is the best scan, and it is good — in a study of 14 obese adults its average agreed with the four-compartment model almost exactly, 41.1% against 41.5% [11]. But the same study found the individual differences between the two “sizable” [11], and you have just seen its 27% on tracking change in lean mass [9]. It is the gold standard the way a good tape measure is: better than the alternatives, not the truth.

A clinical hand-to-foot analyser, the kind in a gym or a physio’s office, missed the four-compartment model by a typical 4.2 points in 82 normal-weight young adults — 2.5 in men, 4.7 in women [10]. That is a typical miss (the study’s “total error”), not the band 19 readings in 20 fall inside, which is wider: in the 2023 test, which reports both, a research-grade analyser with a typical miss of 3.3 points had a band of plus or minus 6.4 [15]. Better than a bathroom scale. Not a decimal place.

Calipers and a tape? A purpose-built equation using three skinfolds and a waist measurement, checked against a five-compartment criterion in 78 adults held back to test it, from a sample aged 18 to 69, put 19 readings in 20 within plus or minus 8.6 points of it [12]. That is with a trained person doing the pinching. In your bathroom, wider.

Put them on that one yardstick, the band 19 readings in 20 fall inside, and you get a rough order rather than a league table, because each figure comes from its own study, its own people and its own reference. DXA: plus or minus 4.2, the study’s own 95% range for predicting the lab model’s figure from a scan, in 14 obese adults [11]. An analyser you grip and stand on: plus or minus 6.4 for the research-grade one, 6.6 to 8.8 for the three consumer ones [15]. The skinfold equation done well: plus or minus 8.6 [12]. The scales you simply stand on: plus or minus 9.3 to 14.7 [15]. Every one of them misses one person’s number by more than a decimal place, and the scale under your sink misses it by the most. Which is why the useful question was never “which is accurate” but “which is consistent enough to compare with itself”.

Where the decimal place came from

In 1985 a research group validated impedance as a lab method for fat-free mass — electrodes on the hand and the foot, so the current crosses the whole body [13]. In 1994 the first consumer body-fat scale went on sale, and to make it a scale you stand on, the electrodes moved to the feet. The current now runs up one leg and down the other, through the legs and lower trunk only, and never reaches the arms or the upper body [1]. That foot-to-foot arrangement is the one whose worst scale could read 14.5 points high; the hand-to-foot arrangement the method was validated with topped out at 6.6 [1].

Then the display printed the estimate to one decimal place, and a band as much as twenty-three points wide became “22.4%”.

Where it lives now: ask Google whether these scales are accurate and its AI Overview answers by citing Healthline, Houston Methodist and a PMC paper — and also a scale manufacturer’s own website, a weight-loss subscription programme, Reddit and YouTube. And “body fat scale” itself is searched 6,600 times a month in the US, with the highest advertiser competition among the related searches in the same pull (from a paid keyword database, pulled 12 September 2026), because the thing people want to know about is the thing being sold.

None of that is aimed at anyone who owns one. The scale is not lying to you; it is estimating, and the display was designed by someone who did not want to print a range.

What a scale is actually good for

This verdict is not “throw it out”, because that is not what the evidence says either.

Weigh yourself on it. The weight is a real measurement, and what the scale weight does and does not tell you is a separate question this site has already taken.

If you use the body-fat number, use it as a yardstick, not a measurement. Same time of day, after the bathroom, before food or water, same scale. Ignore anything under about five points. Look at months, not weeks. Under those conditions a large change is probably real in direction; in a 2023 test over 12 to 16 weeks, the band that 19 in 20 measurements of one person’s change fell inside was plus or minus 3.4 to 5.2 points, depending on the device [15].

Do not chase the number to a target. The ranges those targets come from are softer than the charts imply, and now you know the reading is too. A soft target measured with a soft instrument is not a goal, it is two error bars overlapping.

What this is rated, and what the rating covers

Unsupported. The claim is that the body-fat number on a consumer scale is an accurate measurement of the person standing on it.

It is rated Unsupported because a 2008 test found that a single reading from a foot-to-foot scale could land anywhere from 8.6 points below the reference to 14.5 above on the worst scale, and from about 6 below to 7 above on the best [1], and a 2023 test of ten such scales against a four-compartment model found bands of plus or minus 9.3 to 14.7 points [15]; because the error is systematic rather than random, with a direction that depends on the device and the person [1] [2] [3]; because a drink or a meal shifts the reading by about half a point to a point on its own [5] [7] [16], more if you also stand still for an hour [6], and a day of meals left the estimate unsettled the next morning [14]; and because a study that tested tracking against a four-compartment model, in adults over 65, found individual change in lean mass correctly identified a quarter of the time by impedance, and barely more often by DXA [9], while in the 2023 test the band that 19 in 20 measurements of one person’s change in body fat fell inside was plus or minus 3.4 to 5.2 points, depending on the device [15].

What is not rated: use of these scales on groups, where the 2008 study found them genuinely accurate [1] and the 2023 test found most, though not all, close on average [15]; devices that read through the hands as well as the feet, both the consumer kind with a handle [1] [15] and clinical analysers [10], which had narrower bands in these tests [1] [15] and are a different claim from the one rated here; wearables, on one trial [4]; and what a body-fat percentage means once you have one, which is its own verdict. If what you are really trying to tell apart is fat and loose skin, that is here. How we read a study, and what each tier means, is set out here.