September 1, 2026 · 7 min read · by Quanta Calculator

BMI's Limits: What the Index Measures and What It Can't

BMI is exact arithmetic and a solid population screen, but a poor individual body-fat measure — where the index works, where it fails, and what children use instead

Minimalist geometric illustration of a bathroom scale, measuring tape and bell curve in warm amber tones

Is BMI accurate? As arithmetic, completely — body mass index is weight in kilograms divided by the square of height in metres, and that division carries no error bar. As a verdict on one person's health, no, and it was never built to be one. BMI is a screening index: a cheap, reproducible way to sort large groups of people by body size using nothing but a scale and a tape measure. What it cannot do is measure body fat, say where fat is stored, or tell a muscular frame from a heavy one — which is usually what people asking about its accuracy actually want it to do.

So the honest answer splits in two. Across populations, BMI correlates with directly measured body fat at roughly r = 0.7 in mixed adult samples, and it tracks outcomes clearly enough that the Prospective Studies Collaboration — a 2009 Lancet meta-analysis pooling 894,576 adults from 57 studies — found mortality lowest in the 22.5-to-25 range. For an individual, the same number can describe a rugby prop forward and a sedentary office worker without distinguishing them. The rest of this article covers what the index actually computes, why it took the form it did, and the specific situations — trained muscle, different ancestries, older age, anyone under 20 — where the standard reading gives way.

The computation itself

BMI = weight (kg) ÷ height (m)²

Take a 70 kg adult standing 1.75 m tall. Square the height: 1.75 × 1.75 = 3.0625 m². Divide: 70 ÷ 3.0625 = 22.86 kg/m². Against the WHO adult bands — under 18.5 underweight, 18.5 to under 25 normal weight, 25 to under 30 overweight, 30 and above obese — that lands in the normal range. The imperial version, 703 × pounds ÷ inches², is the identical formula with the unit conversions folded into a single constant. The BMI calculator accepts either unit system and converts internally using the exact definitions of the pound and inch, so metric and imperial entries for the same person agree to two decimal places.

Notice the units of the answer: kilograms per square metre. BMI is not a percentage of anything and not a quantity of fat. It is a density-like ratio of mass to a squared length, and everything the index gets right and wrong follows from that.

A statistician's index, not a physician's

The formula predates its medical career by well over a century. Adolphe Quetelet, the Belgian statistician, introduced it in 1832 while building mathematical descriptions of the "average man" — his subject was the shape of populations, not the health of patients. The name "body mass index" only arrived in 1972, when the physiologist Ancel Keys compared several height-weight ratios against skinfold measurements of body fat in the Journal of Chronic Diseases and found that weight over height squared correlated best. From there the index moved into actuarial tables, into US NIH clinical guidelines in 1998, and into the WHO classification the whole world now quotes.

That lineage explains the design. A screening tool's job is to flag members of a crowd for a closer look, and it can tolerate misclassifying individuals so long as it sorts the crowd roughly right. A diagnostic tool measures the thing itself in the person in front of you. BMI has always been the first kind of instrument, and most arguments about its "accuracy" come from pressing it into service as the second.

Two bodies, one number

Stand two men side by side, both 1.80 m tall and both weighing 95 kg. Height squared is 1.80 × 1.80 = 3.24 m², so each man's BMI is 95 ÷ 3.24 = 29.32 — the top of the overweight band, a fraction below the obesity threshold. Suppose one is a professional rugby prop whose bulk is predominantly muscle and the other has not exercised in a decade. The index cannot separate them, because muscle tissue is roughly 18% denser than fat: an athlete who replaces fat with muscle becomes heavier for the same height and scores a higher BMI while getting leaner. This is why rowers, sprinters, weightlifters and rugby players routinely read as overweight or obese on paper while carrying less fat than the general population.

When body composition is the real question, no index built from weight and height alone can answer it, however carefully it is read. The body fat calculator implements the closest thing to an at-home alternative: the US Navy circumference method, a pair of regression equations fitted by Hodgdon and Beckett in 1984 against hydrostatic underwater weighing in 602 men and 214 women, with typical errors of 3 to 4 percentage points. That error band deserves a hard look — even a purpose-built estimate from neck, waist and hip tape measurements misses by several points for individuals — but at least it estimates the quantity people actually care about, from measurements BMI never takes.

Cut-points that assume a reference population

The 18.5 / 25 / 30 thresholds were derived mostly from European-ancestry cohorts, and they travel imperfectly. A WHO Expert Consultation published in the Lancet in 2004 reviewed evidence that many Asian populations — South Asian, East Asian and Pacific Islander groups among them — develop type 2 diabetes and cardiovascular disease at lower BMI values than European-ancestry populations, and recommended supplementary public-health action points at 23 and 27.5 for those groups; health systems in Singapore, Hong Kong and India have formally adopted lower cut-offs. The index also drifts at the extremes of stature, because a height-squared denominator does not scale perfectly with real body geometry — Quetelet himself noted as much. It weakens again in adults over 65, where age-related muscle loss can hold the number steady while fat rises, and in pregnancy, which the adult categories were never designed to describe.

Children get percentiles, not thresholds

None of the adult numbers apply before age 20 — not approximately, not at all. A BMI of 22 kg/m² classifies a seven-year-old boy as obese yet is a healthy weight for a thirty-year-old woman, because typical adiposity changes continuously as children grow. Pediatric practice therefore replaces fixed cut-points with the CDC's sex-specific BMI-for-age reference: a child's BMI is compared, through the LMS statistical method (Cole and Green, 1992), against reference data for the same sex and the same age in months, yielding a percentile instead of a universal category.

Category	Adults (20+)	Children and teens (2–19)
Underweight	below 18.5 kg/m²	below the 5th percentile
Healthy weight	18.5 to under 25	5th to 84th percentile
Overweight	25 to under 30	85th to 94th percentile
Obesity	30 and above	95th percentile and above

Worked through: a boy of 9 years 6 months — 9 × 12 + 6 = 114 months old — weighs 32 kg at 135 cm. His BMI is 32 ÷ (1.35 × 1.35) = 32 ÷ 1.8225 = 17.56 kg/m². Read against the adult bands, 17.56 would fall below the 18.5 underweight line; read against the CDC reference for 114-month-old boys, it sits at the 73.4th percentile (z-score 0.62) — a healthy weight, higher than about three-quarters of boys his exact age. The gap between those two readings is the entire argument for age-specific references. The children's BMI calculator performs the LMS lookup for any age from 2 through 19 and reports the percentile, the z-score and the category together.

What the number is still for

Used as designed, BMI earns its keep. It costs nothing, requires no equipment beyond a scale and a tape, is perfectly reproducible between clinics and countries, and at population scale it maps onto health outcomes well enough to have anchored decades of epidemiology. For any single person it is one data point — meaningful only alongside waist circumference, blood pressure, blood work and history, and weighing all of that together is a clinician's job, not a calculator's and not this article's. Nothing here diagnoses anything; that division of labour — the tool computes, the clinician interprets — is how every health calculator on Quanta is meant to be read, and this one more than most.

BMI has survived nearly two centuries of criticism because everyone can see exactly what it is: one division, stated in the open, claiming nothing beyond its own units. This article aims for the same property. Each figure in it either recomputes by hand or sits in a source one of the three calculators cites, and a reader who catches one out of place can put the correction in front of the maintainers directly through the contact page.