What BMI is measuring
Weight divided by height squared. It was devised in the 1830s to describe populations, not individuals, and it remains a good screening tool and a poor diagnosis. It cannot distinguish muscle from fat, cannot see where fat is carried, and does not account for bone density or build.
Which is why a rugby forward and a sedentary person of the same height and weight get the same number and are in entirely different health. Waist circumference and waist-to-height ratio track cardiovascular risk more closely, because abdominal fat is the part that matters most.
The thresholds are not universal
The familiar cut-offs — 18.5, 25, 30 — were derived largely from European populations. The WHO recommends lower action points for South and East Asian populations, where cardiometabolic risk rises at a lower BMI; several countries use 23 rather than 25 as the overweight threshold.
They also do not apply to children, who are assessed against age-and-sex percentiles, nor straightforwardly during pregnancy.
What to do with the number
Treat it as one input among several. A BMI outside the usual range is a reason to look further — at waist measurement, blood pressure, blood glucose, activity — not a conclusion in itself. It is a starting point for a conversation with a doctor, and this page is not one.
Frequently asked questions
Is BMI accurate for athletes?
No. Muscle is denser than fat, so a very muscular person often lands in the "overweight" range while carrying little fat. BMI cannot tell the two apart.
Why do some countries use different thresholds?
The standard cut-offs came largely from European data. The WHO recommends lower action points for South and East Asian populations, where risk rises at a lower BMI.
What is a better measure?
Waist-to-height ratio tracks cardiovascular risk more closely, because it reflects abdominal fat specifically. Keeping your waist under half your height is the usual rule of thumb.
Does it work for children?
Not with these thresholds. Children are assessed against age and sex percentiles, which is a different calculation entirely.
Something wrong with this tool, or an idea for it? Tell us