When BMI gets it wrong

A 34-year-old man, four gym sessions a week, BMI 28.2. Every chart calls him overweight. His body-fat percentage, measured on a DXA scan that morning, is 11%. That contradiction is not a bug in his calculator. It’s a feature of BMI.

BMI was invented in the 1830s by Adolphe Quetelet, a Belgian astronomer who was building population statistics for the average man. It does one thing well: track weight against height, across millions of people, in a way that catches population-level shifts. Then doctors started using it on individuals, and the cracks started to show.

TL;DR

• BMI works fine for most sedentary adults of average build. It fails predictably in five specific situations.
• Athletes: muscle weighs more than fat, so trained bodies score “overweight” while being metabolically lean.
• Older adults (≥65): sarcopenia hides under “normal” BMI; the lowest mortality sits in the 25–27.9 range.
• Asian descent: WHO 2004 added separate thresholds (≥23 for increased risk, ≥27.5 for high risk).
• Pregnancy: BMI doesn’t apply; pre-pregnancy BMI guides weight-gain targets instead.
• Children: age- and sex-specific percentiles, not adult cutoffs.

If you fall into one of these groups, treat BMI as a screening flag, not a verdict.

Athletes and muscle-trained adults

Muscle tissue has a density of about 1.06 g/cm³. Fat tissue sits at around 0.9 g/cm³. Pack the same volume of each into your body and the muscle-heavier version weighs more. BMI does not know that.

A useful number to keep in mind: in the Romero-Corral 2008 study (n = 13,601), about half of US adults with a “normal” BMI between 18.5 and 24.9 still met body-fat criteria for obesity. The mirror error runs the other way for trained adults. Powerlifters, rugby players, regular gym-goers in their 20s and 30s often sit at BMI 26–30 with body-fat percentages in single digits. The chart calls this overweight. The metabolic markers and waist circumference disagree.

If you train with weights two or more sessions a week, two numbers matter more than your BMI. The first is your body-fat percentage, ideally measured with DXA, BIA, or skinfolds (in declining accuracy). The second is your waist-to-hip ratio. Either of these will catch fat accumulation that matters and ignore the muscle that doesn’t.

Older adults — the U-shaped curve

Sarcopenia, the gradual loss of skeletal muscle, starts in your thirties and accelerates after sixty. If you maintain the same BMI from 35 to 75 by losing 10 kg of muscle and gaining 10 kg of fat, your chart classification stays “normal” and your actual health profile worsens.

The Winter et al. 2014 meta-analysis (American Journal of Clinical Nutrition) pulled together 32 cohorts covering roughly 197,000 adults over 65. Mortality across the BMI range traced a U: the lowest risk sat in 25–27.9, the bracket that adult charts call “overweight”. Below BMI 22, mortality rose steeply. The signal stayed after adjusting for smoking, illness, and weight loss, suggesting the U is real.

Practical reading: if you are over 65, “underweight” or pushing toward the lower end of “normal” is rarely a goal worth chasing. Preserving muscle, not minimising weight, is the longevity-relevant target. ESPEN 2015 and the GLIM 2019 framework both lean on additional measures such as mid-arm circumference, calf circumference, and grip strength — markers BMI ignores entirely.

Asian descent — different biology, different cutoffs

The 2004 WHO Expert Consultation, published in The Lancet, addressed something Asian researchers and clinicians had been documenting for years: at the same BMI, populations of Chinese, Japanese, Korean, Indian, Vietnamese, Thai, Sri Lankan, and Bangladeshi descent showed higher rates of type 2 diabetes, hypertension, and cardiovascular disease than European populations. The standard 25/30 thresholds were missing risk that needed catching.

The consultation proposed action points rather than new diagnostic categories: BMI ≥23 to flag “increased risk” and ≥27.5 to flag “high risk”. A reading of 26 — comfortably in the standard “overweight” band — sits closer to the standard “obese” zone on the Asian-population chart in terms of cardiometabolic implication.

This applies to ancestry, not residence. A second-generation Bangladeshi-British adult in Manchester carries the same metabolic profile as one in Dhaka. NICE updated UK guidance to reflect this in 2013 (PH46). If you are of Asian descent, ask your clinician to apply the Asian cutoffs, or use a BMI calculator that lets you switch the threshold band.

Pregnancy and post-partum

BMI does not classify a person while they are pregnant. The reason is mechanical: a growing uterus, expanding blood volume, and amniotic fluid all add mass that has nothing to do with the mother’s body composition.

What BMI is used for in pregnancy is your pre-pregnancy reading, which sets the target gestational weight gain. The Institute of Medicine 2009 guidelines (still the active reference in 2026, with the National Academies’ updated review pending) recommend:

Goldstein et al. 2017 in JAMA, a meta-analysis of more than 1.3 million pregnancies, confirmed that gains outside these bands raise the risk of small-for-gestational-age births (when too low) and large-for-gestational-age births and gestational diabetes (when too high). The associations are not subtle.

After delivery, BMI takes time to mean what it used to. Most clinicians wait until at least 12 months post-partum before treating a BMI reading as a stable individual indicator again. Six-week or three-month numbers should not drive decisions.

Children and adolescents

A 12-year-old boy with BMI 22 might sit at the 50th percentile for his age (healthy). His twin brother, born twelve months later in the same body, would land closer to the 75th. Adult thresholds are useless here, because growth itself moves the goalposts.

Paediatric BMI is read against age- and sex-specific percentile curves. The two main reference standards are the CDC 2000 growth charts (commonly used in the United States) and the WHO 2007 references (the international default). Both define the bands the same way:

If you are looking up a child or teenager, use a paediatric BMI calculator, not the adult one. The shape of normal growth makes any single cutoff misleading at this age.

What to use instead

There is no single replacement number. There are better combinations.

The most pragmatic clinical pairing is BMI plus waist circumference. WHO cutoffs for substantial cardiometabolic risk: men >102 cm, women >88 cm. Adding the waist measurement catches central obesity that BMI misses, and adds essentially zero cost.

If you train with weights or are an athlete, body-fat percentage is the more honest number. DXA scans are the most accurate, bioelectrical impedance and skinfolds are cheaper and rougher. For people of Asian descent, switch the BMI cutoff bands rather than abandoning BMI entirely.

In clinical settings, frameworks such as the Edmonton Obesity Staging System layer BMI on top of metabolic, physical, and psychological markers — turning a single number into a clinical picture.

Use BMI as a flag, not a verdict

BMI is fast, free, and directionally useful for most adults of average build. That is exactly what it was designed for. The five situations above don’t make BMI broken — they’re the limits where the tool stops being precise enough to act on alone.

If you want to run your own number, our BMI calculator does it in your browser and shows the WHO classification along with the Asian-population cutoffs. For body-composition context, body-fat percentage and waist-to-hip ratio sit alongside it.