Body fat percentage: every way of measuring it is wrong, differently

Start with a correction of our own

Until this week the body fat calculator on this site was wrong, and not slightly. It was feeding centimetre measurements into a formula whose coefficients require inches, which made men read about 6 percentage points too high and women about 27.

It went unnoticed because of a sanity check. Results above 60 percent were treated as a bad measurement and replaced with "check your measurements", so most women never saw a number at all rather than seeing an obviously absurd one. The guard turned a visible error into a shrug.

It is fixed, and it is worth saying out loud rather than quietly patching, because the rest of this article is about how much confidence these numbers deserve. Ours had a units bug in it for months.

The tape measure method, and what it is actually good at

The US Navy method predicts body fat from your height, waist and neck, plus hips for women. It came from two 1984 Naval Health Research Center reports: 602 men in one, 214 women in the other, with body density measured by underwater weighing.

Against that criterion, in those people, the standard error was 3.5 percentage points for men and 3.7 for women. That is the number the method can honestly claim, and only for a population like the one it was built on.

Against DXA scanning in other groups it does worse and it does so systematically. In army recruits it underestimated body fat by about 6 points in both sexes. In Marines the bias flipped by sex: men read about 2.5 points low, women about 2 points high. A third study found mean bias of 4.8 points low in men and 1.7 in women.

It also fails at the thing people most want it for. Over eight weeks of basic training, DXA detected a real fall of 4 percentage points in women while the tape method showed no change at all. If you are using it to track a twelve-week cut, that is the finding to sit with.

Smart scales, and the hydration claim that is backwards

Bioelectrical impedance scales pass a small current through you and infer composition from the resistance. Three retail models tested against DXA in 134 adults had mean absolute errors of 4.5, 4.9 and 5.9 percentage points. That study was funded by Amazon, which is worth knowing.

Their average bias looks small, but only because the error runs in opposite directions for men and women and cancels out across a group. On you, individually, several points is the expectation.

The hydration warning attached to these scales is real, and the version you usually read has the direction backwards. Being dehydrated makes a scale report LOWER body fat, not higher. In a heat-chamber study, dehydration moved readings from 14.1 to 11.6 percent; overhydration moved them from 13.8 to 14.7. A second study in athletes reproduced both directions while hydrostatic weighing stayed flat.

The practical size of it: a fluid shift of one to three percent of body weight moves the reading by roughly one to two and a half points, and you cannot reliably predict which way. A meal or a large drink is worth about a point on its own.

The ranges everyone quotes have no source

The table of categories you have seen, essential fat at 2 to 5 percent for men and 10 to 13 for women, athletes, fitness, acceptable, obese, comes from the American Council on Exercise. Their own page presents it with no citation, no reference and no derivation. It is a fitness-industry convention that has been copied so widely it reads as a standard.

The essential-fat figures are older still. They trace to a 1950s reference-man and reference-woman concept, propagated through textbooks. There is no modern experimental determination of a minimum survivable body fat percentage, which is unsurprising given what such an experiment would involve.

No medical body publishes healthy body fat ranges. The 2025 Lancet Commission on clinical obesity recommends measuring body fat directly where possible and specifies no percentage cut-offs at all. The common "over 25 percent for men, over 35 for women" thresholds are inherited convention rather than anything derived from outcomes.

What does exist is descriptive population data, and it is sobering next to the fitness table. In national survey data measured by DXA, fat made up roughly 24 percent of body weight in men and 38 percent in women at age 25. The authors present those as reference values, meaning what people are, not what people should be.

Is body fat percentage better than BMI? Genuinely unclear

The fitness world treats this as settled. The literature does not.

For body fat: a BMI of 30 or above misses more than half the people who have excess fat, with sensitivity around 36 percent in men and 49 in women. Body fat percentage carries mortality signal independent of BMI. And in adults aged 20 to 49, one analysis found BMI had no statistically significant relationship with all-cause mortality while body fat did.

Against it: the largest head-to-head comparison, 60,335 people with criterion-standard fat measurement, found models using BMI discriminated cardiovascular mortality significantly better than models using body fat percentage. A separate study found DXA measures added nothing over BMI and waist for cardiometabolic markers. And a collaborative analysis of 58 prospective studies found that once you know someone's blood pressure, lipids and diabetes status, no adiposity measure meaningfully improves cardiovascular prediction at all.

The measure with the best evidence for adding something cheap on top of BMI is not body fat percentage. It is waist-to-height ratio, which improves discrimination by 4 to 5 percent over BMI, and which both the European obesity association and the Lancet Commission now recommend alongside it.

What to do instead of chasing the number

Pick one method and never switch. All of the error above is systematic, which means it is nearly constant for you. A tape measure that reads 5 points low will read 5 points low next month too, so the change between readings is far more trustworthy than either reading.

Measure under the same conditions. Same time of morning, before eating or drinking, and for a scale, the same hydration state as best you can manage. Same-time-different-day precision is about twice as good as different-time-same-day.

Judge over months, not weeks. The tape method failed to detect a real 4-point change across eight weeks. If your method cannot see the change, the absence of movement on it is not evidence.

And measure your waist as well, because that is the number with independent evidence behind it, it takes the same thirty seconds, and it does not depend on a formula at all.

How fast can it change, honestly

Slower than the internet suggests, and the fat share of what you lose is better than people fear. In 140 adults losing about 11 percent of body weight over nine to twelve weeks, measured by DXA before and after, lean tissue made up about a third of the loss in men and a quarter in women. So roughly two-thirds to three-quarters of it was fat, on an ordinary controlled diet.

The "half to one percent of body weight per week" figure is an inference rather than a measured optimum. It comes from a review of natural bodybuilding preparation, which cites two small trials: one testing 0.7 against 1.4 percent per week in 24 elite athletes, and one testing 0.5 against 1.0 kg per week in 15 women. Neither tested the range's lower bound, and neither established it as optimal.

That athlete trial is also the most misreported study in this area. The fast-losing group did not lose lean mass. It failed to gain any, while the slow group gained 2.1 percent, and everyone was doing four resistance sessions a week. Reported as "fast weight loss burns muscle", which is not what it found.