Body Mass Index — weight in kilograms divided by height in meters squared — is calculated on every physical exam, plotted on every growth chart, and cited in every public health statistic about obesity. It is also, by the count of its critics in the scientific literature, an imperfect proxy for the thing it’s supposed to measure, applied inconsistently across populations, and responsible for a meaningful proportion of erroneous clinical classifications.
Understanding what BMI actually measures, and what it doesn’t, requires going back to its origins.
Where BMI came from
The formula weight / height² was proposed by Belgian statistician Adolphe Quetelet in the 1830s — not as a health metric, but as a description of “normal” human proportions. Quetelet was interested in the statistical regularities of human populations. He observed that weight scaled approximately with the square of height in adult populations and used this relationship to define his “Quetelet index.”
For more than a century, this was a curiosity of population statistics, not a clinical tool.
In 1972, American physiologist Ancel Keys published a landmark analysis comparing various weight-for-height indices against body fat percentage measured by densitometry in a large cohort of men. He found that the Quetelet index — which he renamed “Body Mass Index” — correlated better with measured body fat than several competing indices. This paper launched BMI as a health metric.
Keys’ analysis had significant limitations. His sample was entirely male. It was predominantly white and European. It measured body fat by one specific method. But BMI was mathematically simple, required only a scale and a measuring tape, and could be applied at population scale — making it enormously practical.
What BMI actually measures
BMI is not a measure of body fat. It is a weight-for-height ratio. Specifically, it measures the deviation of body mass from what would be expected for a given height, based on the population norm.
A high BMI indicates that a person weighs more than would be expected for their height. It does not indicate why — whether the excess weight is fat, muscle, bone density, or a large frame.
This is the fundamental limitation. A heavyweight bodybuilder with 5% body fat has a BMI in the “obese” range. A sedentary person with very low muscle mass and high fat percentage may have a BMI in the “normal” range. BMI classifies both incorrectly by any measure that actually tracks metabolic risk.
In population-level epidemiology — where the goal is to describe averages across thousands of people — BMI works reasonably well because, on average, high BMI correlates with high body fat at the population level. The correlations break down at the individual level, particularly at the extremes of body composition.
The population calibration problem
Keys’ original BMI analysis was conducted in white European and American men. The cutoffs that became clinical standards (underweight <18.5, normal 18.5–24.9, overweight 25.0–29.9, obese ≥30.0) were calibrated to this population.
Subsequent research found that these cutoffs are poorly calibrated for other populations:
East Asian populations: At the same BMI, East Asian populations have higher body fat percentages and higher metabolic risk than European populations. The WHO convened an expert consultation in 2004 that recommended lower BMI cutoffs for Asian populations — proposed thresholds for “increased risk” at BMI ≥ 23.0 and “high risk” at BMI ≥ 27.5, compared to European thresholds of 25.0 and 30.0. Clinical practice in many Asian countries now uses these lower cutoffs.
African and African-American populations: Some research suggests that African-origin populations have lower body fat percentage at the same BMI compared to European populations, potentially leading to over-classification of risk. This is an area of active debate in the literature.
Children and adolescents: Adult BMI categories are meaningless for children, because expected BMI changes with age and sex. Children are assessed using BMI-for-age percentiles — a child’s BMI compared to the distribution for their age and sex, using growth charts calibrated to the relevant population.
Older adults: In adults over 65, the relationship between BMI and mortality is non-linear and sometimes reversed from the pattern in younger adults — slightly higher BMI is associated with better survival outcomes in some older populations, possibly because BMI in older adults reflects lower risk of malnutrition.
What anthropometry beyond BMI captures
BMI compresses all weight-height variation into a single number. The 130+ dimensions of a full body profile describe the actual geometry of the body with specificity that BMI cannot approach.
Waist circumference is a better predictor of cardiometabolic risk than BMI in most populations studied. High waist circumference (central adiposity) is associated with higher metabolic risk independent of total body weight. This is why waist circumference has been added to risk assessment protocols in most cardiovascular medicine guidelines, alongside or instead of BMI.
Waist-to-hip ratio further refines this: high waist-to-hip ratio (centralized fat distribution, higher “apple shape”) is associated with different risk than high BMI with low waist-to-hip ratio (more peripheral fat, “pear shape”).
Waist-to-height ratio has been proposed as a single-number alternative to BMI that captures central adiposity. The simple rule “keep your waist circumference less than half your height” has been shown to correlate with cardiovascular risk better than BMI in several large studies.
Lean body mass estimates from skinfold measurements or bio-electrical impedance give a more accurate picture of body composition than weight alone — distinguishing muscle from fat, which BMI cannot.
BMI in digital health applications
For digital health applications — fitness apps, wellness platforms, health monitoring devices — BMI is often computed and displayed because it’s simple and well-understood by users. The question is whether it’s the right metric for the application’s purpose.
For population-level health risk communication in an app, BMI-for-age (for pediatric users) or BMI with appropriate population-specific context (for adult users) is reasonable — with clear communication that it’s a rough screening metric, not a diagnostic result.
For fitness tracking, body composition metrics (fat percentage, lean mass) are more meaningful than BMI, because a fitness program that increases muscle mass improves health while potentially increasing BMI. Apps that track only BMI will show counterproductive changes for users who successfully build muscle.
For clinical applications, BMI alone is insufficient without contextual measurements. Waist circumference adds minimal effort and substantially more information.
For product sizing, BMI is nearly useless — the same BMI can correspond to vastly different body shapes and dimensions. Full dimensional prediction (chest, waist, hip, shoulder) is required to support any sizing decision.
The 2023 American Medical Association position
In 2023, the American Medical Association adopted new guidance on BMI as a diagnostic measure, acknowledging explicitly that BMI is an imperfect measure of body fat and health risk, has an “historical harm” component due to its use in racially discriminatory medical practices, and should not be used alone to make diagnostic or treatment decisions.
The AMA recommended that BMI be used alongside other measures including “visceral fat, body adiposity index, body composition, relative fat mass, waist circumference and genetic/metabolic factors.”
This represents a significant shift in the clinical consensus after decades of BMI dominance in clinical guidelines. The direction is toward multi-dimensional body measurement rather than a single weight-height ratio.
BMI persists because it’s simple, requires no special equipment, and has decades of population-level data behind it. These are real advantages for certain applications. But the scientific consensus is increasingly clear that it should be used with awareness of its limitations — and that, wherever possible, more specific body composition or dimensional data produces better clinical and design decisions.
The future of body measurement in health applications is almost certainly multi-dimensional: waist circumference, body composition estimates, and eventually full dimensional profiles as measurement technology makes them more accessible. BMI will likely remain in use as a screening tool long after better tools are widely available — simply because of the infrastructure built around it.