Dual-energy X-ray absorptiometry (DXA) is a valid technique for measuring the fat, bone and lean (muscle, organs and water) masses of the body. We evaluated relationships of BMI (kg/m2) with independent measurements of fat and lean masses using DXA in 226 adult volunteers. The evaluation was an application of a general approach to compositional data which has not previously been used for describing body composition. Using traditional regression analyses, when lean mass was held constant, BMI varied with fat mass (men r 0.75, P < 0.05; women r 0.85, P < 0.05); when fat mass was held constant, BMI varied with lean mass (men r 0.63, P < 0.05; women r 0.47, P < 0.05). In contrast, a regression model for compositional data revealed that BMI was: (a) strongly associated with log fat mass in both sexes (b1 4.86, P < 0.001 for all women and b1 5.96, P < 0.001 for all men); (b) not associated with bone mass, except in older men; (c) related to lean mass in women but not in men (b3 -4.04, P < 0.001 for all women and b1 -2.59, P < 0.15 for all men). Women with higher BMI tended to have more fat mass and more lean mass than women with lower BMI. Men with higher BMI had more fat mass but similar lean mass to men with lower BMI. Investigators need to be alert to the inaccuracy of BMI to assign a fatness risk factor to individuals, especially among women.