Body Mass Index Explained: The Formula, the Cut-Offs, and the Limits

Body mass index is a screening arithmetic at population scale, not a diagnostic for a single body. The formula itself has not changed since Adolphe Quetelet devised it in 1832: weight in kilograms divided by height in metres squared. BMI explained as a health metric is the most common mistake newcomers make. The number was never designed to tell you how much fat you carry or whether you are metabolically healthy. It was designed to spot statistical shifts in large groups. Here you get the formula, the provenance of every cut-off, the groups for whom the standard thresholds misclassify risk, and the measurements that predict health outcomes better than BMI alone. The site sells nothing, prescribes nothing, and will not tell you what to weigh. It will tell you what the number actually means and what it does not.

• Formula: weight (kg) / height (m)². Imperial version: [weight (lb) / height (in)²] × 703.
• Originator: Adolphe Quetelet, 1832, as the Quetelet Index.
• Renamed To BMI: Ancel Keys, 1972 study 'Indices of relative weight and obesity', 7,424 men across five cohorts.
• WHO Adult Cut-Offs (1995/2000): Underweight <18.5, normal 18.5-24.9, overweight 25.0-29.9, obese class I 30.0-34.9, obese class II 35.0-39.9, obese class III ≥40.0 kg/m².
• WHO Asian Cut-Offs (2004): Additional action points at 23.0 kg/m² (increased risk) and 27.5 kg/m² (high risk), retaining standard cut-offs for international comparison.
• Paediatric BMI (US): Plotted on CDC 2000 growth charts, age- and sex-specific percentiles. Underweight <5th, healthy 5th, <85th, overweight 85th, <95th, obesity ≥95th, severe obesity ≥120% of 95th percentile or ≥35 kg/m².

BMI measures weight relative to height squared. It does not measure body fat proportion, lean mass, bone density, or fat distribution. The correlation between BMI and DXA-measured body fat proportion is moderate at population level, with r values between 0.60 and 0.80 depending on sex, age, and ethnicity. That leaves a wide spread of values at any given BMI. A person with a BMI of 27 could have 18% body fat or 35% body fat. The formula cannot distinguish them because weight includes everything: muscle, bone, organs, water, and fat.

Adolphe Quetelet was a Belgian mathematician and statistician, not a physician. He derived the Quetelet Index from studies of French and Scottish conscripts in the 1830s, looking for a description of the 'average man.' The index worked for that purpose: in a large group, weight scales roughly with the square of height. Keys renamed it BMI in 1972 after finding it correlated better with body fat measured by skinfolds and densitometry than other weight-height indices did. But Keys studied only men, and his sample of 7,424 came from five cohorts of railroad workers, middle-aged men, and military conscripts. Not a representative slice of humanity.

The thresholds on every app, scale, and doctor's chart come from the World Health Organization Technical Report Series 854, published in 1995, and refined in the 2000 report Obesity: Preventing and Managing the Global Epidemic. The committee reviewed observational data on morbidity and deaths, primarily from US and European groups. They chose the cut-points where risk of chronic disease began to rise detectably in those epidemiological datasets. There was no physiological threshold, no blood marker, no imaging scan, behind the number 30.0. It was the point at which the curve of cardiovascular disease and type 2 diabetes started to climb in the white, mostly European-descended groups available at the time.

The underweight boundary, 18.5, was set where deaths from infectious disease and wasting conditions increased. The normal range, 18.5 to 24.9, was the band where the pooled death risk was lowest in the data the committee had. That J-shaped relationship between BMI and all-cause deaths was already visible in the 1990s and was confirmed in 2016 by the Global BMI Mortality Collaboration, which pooled prospective studies with 10.6 million participants. The nadir of the curve sits between 20 and 25 kg/m². Each 5 kg/m² increase above 25 was associated with roughly 30% higher all-cause deaths. But the shape of that curve differs by group, by age, and by underlying health status. The cut-offs are a compromise, not a law.

The 2004 Asian Consultation

In 2004, the WHO convened a consultation after evidence from ten Asian countries showed that type 2 diabetes and cardiovascular disease risk increased at lower BMI in Asian groups. The resulting report, Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies, added two action points: 23.0 kg/m² for increased risk, and 27.5 kg/m² for high risk. The committee retained the standard WHO cut-offs for international comparison but advised countries to adopt the lower thresholds for their own public health surveillance. Several jurisdictions did: India, Singapore, Japan (where the Japan Society for the Study of Obesity published its own cut-offs), and the WHO Western Pacific Region. A BMI of 25 in a person of South Asian ancestry carries a metabolic risk profile closer to a BMI of 30 in a white European. At a given BMI, South Asians carry a higher body fat proportion.

WHO, CDC, and NICE all state explicitly that BMI is a screening tool for population-level surveillance, not a diagnostic for a single person. Screening identifies people who may need further assessment. Diagnosis identifies a condition. BMI is validated for the former and not for the latter in every major clinical guideline. The distinction matters. The sensitivity and specificity of BMI for detecting excess adiposity are not good enough for individual decisions.

Using DXA-measured body fat proportion as the reference (≥25% in men and ≥30% in women), the sensitivity of BMI (≥30) for detecting obesity is roughly 50%, with a range of 30% to 70% depending on the group. That means about half of people with excess body fat are missed by the standard BMI cut-off. The false negative rate is especially high in older adults. Up to 50% of those with excess body fat have a normal BMI. Specificity is better, around 90% to 95%. Most people who are not obese by body fat proportion are correctly identified. But the false positive rate matters too. Athletes and muscular people routinely have a BMI of 30 or above while carrying less than 20% body fat in men and less than 30% in women. The correlation between BMI and body fat in Black groups runs lower than in white groups at the same BMI. Standard cut-offs overestimate adiposity in some groups and underestimate it in others.

Metric	What It Measures	Population Cut-Off Example	Individual Diagnostic Use
BMI (kg/m²)	Weight relative to height squared	≥30 for obesity (WHO 1995/2000)	No; screening only
Body fat proportion (%)	Adipose tissue fraction of total mass	Men ≥25%, women ≥30% (method-dependent)	Reference standard with no universal cut-off across methods
Waist circumference (cm)	Abdominal girth, surrogate for central fat	Men >102 cm, women >88 cm (NIH/NHLBI 1998)	Protocol-dependent; self-measurement unreliable
Waist-to-height ratio (dimensionless)	Waist scaled to stature	0.5 (NICE CG189 2022)	Population heuristic, not a personal risk score
Visceral fat area (cm²)	Fat around internal organs at L4-L5	≥100 cm² (Japanese criteria, 2002)	Requires CT or MRI; no universal clinical cut-off

At population scale, BMI is cheap, fast, and good enough as a proxy for adiposity. It correlates with disease risk. It tracks changes in a group over time. It requires only a stadiometer and a scale. Public health agencies use it to allocate resources and to monitor trends because it works at the level of hundreds of thousands of people. The WHO cut-offs have been stable for three decades, which makes them useful for cross-national comparisons even if the risk associated with a given BMI differs across countries.

BMI also predicts risk of type 2 diabetes, cardiovascular disease, and all-cause deaths in large observational studies. Each 1 kg/m² increase in BMI is associated with roughly 20% to 25% higher risk of type 2 diabetes (the exact figure varies by ethnicity). Each 5 kg/m² increase above 25 is associated with about 30% higher coronary heart disease risk, according to the Emerging Risk Factors Collaboration of 2011, which pooled 58 prospective studies with 221,934 participants. For all-cause deaths, the Global BMI Mortality Collaboration found that each 5 kg/m² increase above 25 carried about 30% higher risk. Those are real signals that would be missed if BMI were not measured at all.

What BMI Gets Wrong

BMI gets wrong everything about a single person. It cannot distinguish muscle from fat. It cannot identify where fat is stored. Subcutaneous versus visceral adipose tissue, and visceral fat is the metabolically dangerous compartment. It cannot detect normal-weight obesity: a person with a BMI in the normal range but excess body fat by DXA. The TOFI phenotype, thin outside, fat inside, affects an estimated 20% to 30% of normal-weight adults, who have elevated visceral fat despite a normal BMI. BMI also misclassifies risk by ethnicity. At the same BMI, Black groups have lower body fat proportion, while South Asians have higher body fat proportion and higher cardiometabolic risk. The correlation between BMI and body fat in South Asians is higher than in Europeans. Using the same cut-off for both groups guarantees systematic misclassification.

Self-reported height and weight, which many people rely on for online BMI calculators, systematically bias the number downward. Weight is under-reported, height is over-reported. The gap is larger in heavier people. The average error in adult groups is 0.5 to 1.5 kg/m² downward. In a single person, that can shift someone from the overweight band into the normal band, or from obese class I into the overweight band. That bias matters when BMI is used as a screening tool in settings where measured data are not collected.

If you have a BMI number and you want to know whether it accurately describes your body composition or your health risk, you need additional measurements. Get your waist circumference measured with a tape at the midpoint between the lowest rib and the top of the iliac crest, at the end of a normal expiration. The NIH/NHLBI thresholds, greater than 102 cm for men, greater than 88 cm for women, were published in 1998 based on increased cardiometabolic risk in white groups. The International Diabetes Federation proposed lower thresholds in 2005 for some ethnic groups: men 94 cm, women 80 cm for Europids; men 90 cm, women 80 cm for South Asian, Chinese, and Japanese groups. But the protocol matters. The WHO and NIH use different landmarks, and self-measured waist circumference is unreliable. Errors commonly exceed 3 to 5 cm.

Waist-to-height ratio, which NICE CG189 introduced in 2022 as a practical estimate of central adiposity, removes the need for sex-specific cut-offs. Divide your waist circumference by your height in the same units. Keep the result below 0.5. The evidence base is a meta-analysis of 31 studies with about 300,000 participants, published by Ashwell in 2012. It found that waist-to-height ratio outperformed both BMI and waist circumference for detecting cardiometabolic risk. It is still a population heuristic, not a personal risk score. But it adds information beyond BMI alone and does it with one tape measure and one ratio.

Body fat proportion measured by DXA is the closest thing to a reference standard for individual body composition, with an error of about 1% to 3% compared to the four-compartment model. DXA is expensive. It is not a routine screening tool, and the healthy range depends on the measurement method. Gallagher and colleagues published ranges in 2000 from a cross-sectional study of 1,626 adults using DXA: for men, 10% to 20% at ages 20 to 39, 11% to 22% at ages 40 to 59, 13% to 25% at ages 60 to 79; for women, 20% to 32% at ages 20 to 39, 22% to 34% at ages 40 to 59, 24% to 36% at ages 60 to 79. Those are descriptive ranges from one study, not clinical cut-offs. No universal healthy-range standard exists across all measurement methods and groups. Bioelectrical impedance analysis, the method used by most consumer scales, carries individual errors of 3 to 8 percentage points for body fat proportion compared to DXA. It is highly sensitive to hydration status. A single reading taken after a meal, after exercise, or on a dehydrated morning is noise, not signal.

BMI captures nothing that waist circumference, waist-to-height ratio, DXA-measured body fat proportion, or visceral fat area do not capture better. What BMI has is ubiquity. It is the metric on every doctor's chart, every health insurance form, every population survey. It is the number most people have seen and the number most people remember. That does not make it the right number for an individual decision. The question is not what BMI measures. The question is whether it measures it well enough for a given purpose. For population surveillance, yes. For individual clinical screening, it is one input among several. For a personal health verdict, it is not enough.

Cardiorespiratory Fitness and the Obesity Paradox

Cardiorespiratory fitness, measured as VO₂max (millilitres of oxygen per kilogram of body mass per minute), stratifies death risk within every BMI category. A 2009 meta-analysis by Kodama of 33 studies with 102,980 participants found that high cardiorespiratory fitness was associated with 45% to 50% lower all-cause deaths compared to low fitness, independent of BMI. A 2014 meta-analysis by Barry of 10 studies with about 100,000 participants found that fit individuals with obesity had lower all-cause deaths than unfit normal-weight individuals. That finding overturns the simple assumption that a high BMI is always worse than a low one. A person's fitness level, measured by an exercise test, predicts their health outcomes better than their BMI does.

The obesity paradox, the epidemiological observation that in certain disease cohorts (heart failure or chronic kidney disease) higher BMI is associated with fewer deaths, is not a reason to gain weight. It is a statistical pattern confounded by reverse causation: illness causes weight loss, so the sickest people have the lowest BMI. Smoking also reduces body weight and increases deaths. Mendelian randomisation studies, which use genetic variants to estimate causal effects, suggest that the causal relationship between higher adiposity and deaths is more linear than the observed J-curve implies. The paradox is a caution about observational data, not a biological mechanism.

Older adults lose muscle mass and bone density while gaining fat, often without a change in weight. A normal BMI in a 70-year-old can conceal excess body fat and sarcopenia. The false negative rate for excess body fat in older adults reaches 50% in some studies. Athletes and bodybuilders have high muscle mass and low body fat. Their BMI lands in the overweight or obese range despite body fat proportions well below the threshold for metabolic risk. A male athlete with 12% body fat and a BMI of 28 is not overfat, but the BMI number says overweight. The standard cut-offs were not designed for people with above-average muscle mass, because the reference groups did not include them.

People of Asian, South Asian, and East Asian ancestry carry higher body fat at a given BMI than white Europeans. The WHO Asian cut-offs of 23 and 27.5 are an attempt to correct that. Many national guidelines still use the standard thresholds, producing systematic under-detection of risk. Normal-weight obesity, defined as a normal BMI with body fat proportion above the healthy range, affects a substantial minority. These individuals are invisible to BMI screening but carry elevated cardiometabolic risk. The TOFI phenotype, thin outside, fat inside, is detectable only by direct body composition measurement. Waist circumference and waist-to-height ratio catch some of them, because central adiposity is the driver of risk. Neither measure is routinely collected in primary care. If you rely only on BMI to assess your health, you are missing the people who need intervention most: the ones whose weight looks fine but whose body composition does not.

Waist-to-height ratio, as NICE CG189 recommends, is the cheapest and most practical improvement. The 0.5 threshold works across sex and most ethnic groups. The measurement requires only a tape measure. A 2012 meta-analysis of 31 studies by Ashwell found that waist-to-height ratio outperformed both BMI and waist circumference for detecting cardiometabolic risk across 300,000 participants. The second-best addition is cardiorespiratory fitness. It requires an exercise test but stratifies risk within every BMI category. Blood pressure, measured according to the 2017 ACC/AHA guidelines, adds cardiovascular and renal risk information independent of BMI. Each 1 kg/m² increase in BMI is associated with roughly a 1 to 1.5 mmHg increase in systolic blood pressure. The variability is wide enough that knowing someone's blood pressure adds information beyond knowing their BMI.

Visceral fat measurement by CT or MRI is the most precise. It is not a screening tool. The Japanese threshold of 100 cm² at L4-L5 has been used since 2002, but no universal clinical cut-off exists. The proportion of normal-weight adults with elevated visceral fat, the TOFI phenotype, is high enough that assuming a normal BMI means a normal metabolic risk profile is a mistake. DXA-measured body fat proportion, despite being the closest to a reference standard, has no universal healthy-range cut-off across measurement technologies. The Gallagher ranges from 2000 are age- and sex-specific but derived from one study using DXA. They are not a clinical threshold.

WHO, CDC, NICE, and the US Preventive Services Task Force all classify BMI as a screening tool. NICE CG189 states that BMI should be used as a practical estimate of overweight and obesity for population health, not as a definitive measure of body fat for individuals. The document explicitly recommends waist-to-height ratio as a central adiposity marker for adults with a BMI below 35. The NIH/NHLBI 1998 clinical guidelines on overweight and obesity included waist circumference as a risk modifier within each BMI category, acknowledging that the same BMI carries different risk depending on fat distribution.

The distinction between screening and diagnosis is not a technicality. A screening test trades individual accuracy for population coverage. It is designed to flag people who might need further assessment, not to deliver a verdict. The 50% sensitivity of BMI for detecting excess body fat means that for every person correctly flagged, another person with the same body composition is missed. That is acceptable for a population survey that tracks trends over time. It is not acceptable for a clinical decision about a single person. If you are using BMI alone to decide whether someone needs metabolic testing, dietary intervention, or medical treatment, you are misusing the tool.

1832: Adolphe Quetelet publishes the Quetelet Index, a statistical description of the average man, based on conscript data. No health application is intended.

1972: Ancel Keys publishes 'Indices of relative weight and obesity', a study of 7,424 men across five cohorts, finding that the Quetelet Index correlates better with body fat measured by skinfolds and densitometry than other weight-height indices. Keys renames it the Body Mass Index.

1995: The WHO publishes Technical Report Series 854, Physical Status: The Use and Interpretation of Anthropometry, establishing the adult BMI cut-offs: underweight below 18.5, normal 18.5 to 24.9, overweight 25.0 and above, obese 30.0 and above. The evidence base is observational data on morbidity and deaths from US and European groups.

2000: The WHO refines the classification in Obesity: Preventing and Managing the Global Epidemic, adding the three obesity classes and confirming the thresholds.

2004: The WHO Western Pacific Regional Office publishes Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies, adding action points at 23.0 and 27.5 for Asian groups while retaining standard cut-offs for international comparisons.

2022: NICE CG189 introduces waist-to-height ratio as a central adiposity marker for clinical use alongside BMI, with a 0.5 threshold for adults with a BMI below 35.

The thresholds have not changed since 2000 for the general adult population. Any revision would require a new WHO expert consultation and a systematic review of the evidence. The durability of the cut-offs is a feature: it allows cross-national and cross-temporal comparisons. It is also a limitation. The evidence base that set them was drawn from groups that are not representative of the world's ethnic diversity.

Measure your waist circumference and height using a tape measure, then calculate your waist-to-height ratio. If the result is below 0.5, your central adiposity is within the population heuristic that NICE CG189 recommends. If it is above 0.5, discuss it with a healthcare provider. Do not use BMI alone. Do not let a single number from an app or a scale tell you whether you are healthy. The number is a screening tool, not a verdict. The next step is not a diet plan or a workout schedule: the site does not prescribe those. Get a measured waist circumference and a measured waist-to-height ratio, then seek a clinical consultation if you have concerns.