Short answer: on average, surprisingly close; for you specifically, not close enough to trust the number. Bosy-Westphal and colleagues (2008) tested four consumer bioimpedance devices against DXA and whole-body MRI in 106 adults. Three of the four had a mean error under 1.5 percentage points of body fat, which is excellent for a group. But the limits of agreement, the range within which an individual's reading could fall, ran from about −6.6 to +4.6 points for the best device and −14.5 to +8.6 points for the worst foot-to-foot scale. A reading of 25% could mean 18% or 30%. The scale is measuring electrical resistance through your legs and feeding it to an equation; the equation is right on average because it was fitted to a population, and wrong for individuals because you are not a population.
This is the same distinction as the calorie number on a smartwatch: weight is measured, body fat is inferred, and the two are shown in the same font.
What the scale is actually doing
A body-fat scale passes a small alternating current between the electrodes under your feet and measures the impedance. Lean tissue is mostly water and conducts well; fat conducts poorly. From the impedance, your height, weight, age and sex, and a regression equation developed on some reference group, the device estimates total body water, then fat-free mass, then fat as the remainder.
Three things follow from that description, and the European clinical guidelines on bioimpedance by Kyle and colleagues (2004) spell out all of them.
It never measures fat. It measures how easily current passes through the path it takes. Everything after that is a model.
On a foot-to-foot scale, the current goes up one leg and down the other. It does not pass through the trunk, where a large share of adult fat sits. The equation assumes your trunk is like the trunks in the reference group. If you carry fat differently from that group, the equation does not know.
Anything that changes body water changes the reading. Hydration status, a meal, a workout, the time of day, alcohol the night before, a menstrual cycle phase, a hot bath. The guidelines list all of these as things to standardise before a clinical measurement, and none of them are standardised when you step on the scale after breakfast.
The validation numbers
| Device type | Sample | Mean bias vs DXA (% body fat) | 95% limits of agreement | Source |
|---|---|---|---|---|
| Tetrapolar (hand-and-foot) consumer device | 106 adults | <1.5 points | −6.6 to +4.6 | Bosy-Westphal 2008 |
| Two foot-to-foot consumer scales | Same | <1.5 points | Wider than tetrapolar | Same |
| Worst foot-to-foot consumer scale | Same | ~3 points | −14.5 to +8.6 | Same |
| Multifrequency laboratory BIA | Large healthy adult sample | Small at group level | Individual error of several points, larger at the extremes of body fat | Sun 2005 |
Read the limits-of-agreement column, not the bias column. A bias under 1.5 points means that if you weighed a hundred people the average error would be small, which is why bioimpedance is useful in epidemiology, as Dehghan and Merchant (2008) argue. The limits of agreement are what happen to one person, and they are wide. The authors' own recommendation was that the tetrapolar arrangement, with hand electrodes as well as foot electrodes so the current crosses the trunk, should be preferred for individual use, while foot-to-foot devices are suited to group-level research.
Limits of agreement ranged from −6.59 to 4.61 percent fat mass for the tetrapolar device and were as wide as −14.54 to 8.58 percent fat mass for one bipolar device.
Bosy-Westphal et al, Obesity Facts, 2008
The pattern that recurs across the validation literature is also worth knowing: consumer scales tend to overestimate fat in lean people and underestimate it in people with obesity. The equations pull everyone toward the middle of the group they were built on. If you are at either end, expect the bias to be larger and in a predictable direction.
What moves the number day to day
The second problem is noise. Even if the scale were unbiased for you, the reading swings with body water.
| Condition | Direction of body-fat reading | Why |
|---|---|---|
| First thing in the morning, before drinking | Higher | Overnight dehydration raises impedance |
| After a large glass of water or a meal | Lower, then variable | Water in the gut is not yet distributed; the model treats it as lean mass |
| After exercise | Usually lower, sometimes higher | Sweat loss raises impedance; blood flow to the legs lowers it; the two compete |
| After alcohol | Higher next morning | Dehydration |
| Warm feet, warm room | Lower | Better skin contact and blood flow reduce impedance |
| Cold feet, dry skin | Higher | Poor electrode contact |
| Menstrual cycle, luteal phase | Higher | Water retention shifts the model |
| Different scale, same person | Different | Different equation, different reference group |
The practical consequence is that day-to-day changes of one to three points are noise. They are not fat lost or gained; nobody gains two points of body fat overnight. Anyone who has watched the number go up after a good week of training has met this, and usually blamed the training.
What the number is good for
A body-fat scale is a poor instrument for the question "what is my body fat percentage" and a reasonable one for "is it going up or down over months", provided the conditions are held constant. That is the same rule as for every consumer sensor on this site, from HRV to sleep tracking: the trend, measured consistently, contains information; the absolute value mostly does not.
The conditions that make the trend usable:
Same time, same state. Morning, after using the bathroom, before drinking or eating, bare feet, same scale. Every time.
Weekly average, not daily reading. Take it daily if you like, but plot the seven-day average. The daily swings cancel.
Months, not weeks. A real change of two points of body fat in a 75 kg person is about 1.5 kg of fat, which for most people takes six to ten weeks. Anything faster than that on the scale is water.
Weight and waist alongside. Weight is measured accurately by the same device. A tape around the waist at the navel is cheap, has its own error, and does not depend on your hydration. If all three move the same way over two months, something is happening. If only the fat percentage moves, it is the model.
What would be more accurate
DXA. The reference method in most of these studies. A low-dose X-ray scan that separates bone, lean and fat with an error of a point or two. Available at some clinics and universities; costs money; the right answer if you need a number.
Hand-and-foot bioimpedance. The tetrapolar layout crosses the trunk and had the narrowest limits in the validation study. If buying a scale, this is the specification that matters more than the app.
Skinfold callipers, by someone trained. Operator-dependent and good in practised hands, poor otherwise.
Waist circumference. Not body fat, but a better predictor of the health outcomes people care about than fat percentage is, with a tape measure and no model.
Photographs and a mirror, monthly. Crude, free, and not fooled by hydration.
What to actually do
If you own one: keep using it for weight, and treat the fat number as a weekly trend under identical morning conditions. Ignore any single reading. Ignore the "muscle mass" and "visceral fat" numbers, which are derived from the same impedance by further equations and have not been validated to a standard worth quoting.
If you are buying one: hand-and-foot electrodes over foot-only. The difference in the limits of agreement is the difference between a wide guess and a very wide guess.
If you want the actual number: a DXA scan, once, and then use the scale's trend from that baseline.
If the number is upsetting you: put the tape measure and the mirror in charge and give the scale a rest. It is showing you body water with a confident decimal point.
Questions people ask
How accurate are body fat scales? On average, within about 1.5 percentage points of a DXA scan for the better consumer devices. For an individual, the reading can be off by 5 to 10 points or more: limits of agreement in the main validation study ran from −6.6 to +4.6 points for the best device and −14.5 to +8.6 for the worst.
Are smart scales accurate for body fat? For weight, yes. For body fat, they are accurate for groups and unreliable for individuals, because the number comes from an equation fitted to a reference population rather than from a measurement of your fat.
Why does my body fat scale change every day? Because it measures electrical impedance, which tracks body water. Hydration, meals, exercise, alcohol, room temperature and foot contact all move the reading by one to three points from day to day without any change in fat.
Do body fat scales overestimate or underestimate? Both, depending on where you sit. Consumer scales tend to overestimate fat in lean people and underestimate it in people with obesity, pulling everyone toward the average of the population the equation was built on.
When is the best time to use a body fat scale? First thing in the morning, after using the bathroom, before eating or drinking, with bare, clean feet, at the same time each day. Consistency matters more than the specific time.
Are hand-and-foot scales more accurate than foot-only? Yes, meaningfully. The current crosses the trunk, where much adult fat is stored, and the tetrapolar device had the narrowest individual error in the validation study. The authors recommended it for individual use over foot-to-foot scales.
Can a body fat scale track progress? As a weekly-averaged trend under identical conditions over months, yes. As a daily readout, no. Pair it with weight and waist circumference; if all three move together, the change is real.
Is DXA more accurate than a body fat scale? Substantially. DXA is the reference method most scales are validated against, with individual error of a point or two. It is the option if the actual number matters.
Are the muscle mass and visceral fat numbers on smart scales accurate? They are further estimates derived from the same impedance reading and have less validation behind them than the fat percentage. Treat them as decoration.
Is body fat percentage even the right number? For health risk, waist circumference and waist-to-height ratio predict outcomes at least as well and do not depend on a model. For appearance and training progress, photographs and the mirror are hard to fool.
This article covers consumer body-composition devices for general information and is not medical advice. Body fat scales pass a small electrical current and are not recommended for people with implanted cardiac devices; check the manufacturer's guidance.
References
- Bosy-Westphal, A., Later, W., Hitze, B., et al. (2008). Accuracy of Bioelectrical Impedance Consumer Devices for Measurement of Body Composition in Comparison to Whole Body Magnetic Resonance Imaging and Dual X-Ray Absorptiometry. Obesity Facts, 1(6), 319–324. doi:10.1159/000176061
- Kyle, U.G., Bosaeus, I., De Lorenzo, A.D., et al. (2004). Bioelectrical impedance analysis—part I: review of principles and methods. Clinical Nutrition, 23(5), 1226–1243. doi:10.1016/j.clnu.2004.06.004
- Kyle, U.G., Bosaeus, I., De Lorenzo, A.D., et al. (2004). Bioelectrical impedance analysis—part II: utilization in clinical practice. Clinical Nutrition, 23(6), 1430–1453. doi:10.1016/j.clnu.2004.09.012
- Dehghan, M., & Merchant, A.T. (2008). Is bioelectrical impedance accurate for use in large epidemiological studies? Nutrition Journal, 7, 26. doi:10.1186/1475-2891-7-26
- Sun, G., French, C.R., Martin, G.R., et al. (2005). Comparison of multifrequency bioelectrical impedance analysis with dual-energy X-ray absorptiometry for assessment of percentage body fat in a large, healthy population. The American Journal of Clinical Nutrition, 81(1), 74–78. doi:10.1093/ajcn/81.1.74
The weekly readout
One email each Thursday: what we tested, which claim collapsed under a closer look, and the one number worth paying attention to.