Body composition was measured in twenty young females aged 19−27 years and eighteen elderly females, aged 65−78 years by densitometry (underwater weighing), deuterium oxide dilution and dual-energy X-ray absorptiometry (DXA). From body weight, bone-mineral content, total body water and body density, percentage body fat (BF%) was calculated using a four-compartment model. BF% abtained by this four-compartment model was regarded as a reference method and BF% obtained by the single methods were compared with this value. Differences in BF% from the four-compartment model minus the single methods were 2·1 (SD 1·2) for densitometry, 3·1 (SD 1·8) for DXA and -0·6 (SD 0·9) BF% for deuterium oxide dilution in the young women. In the elderly women these values were -0·6 (SD 2·3), 5·3 (SD 3·8) and 0·7 (SD 2·2) BF%. When a three-compartment model (calculated from body density and total body water) was compared with the four-compartment model, the bias was 0·4 (SD 0·3) BF% in the young and 0·0 (SD 0·3) BF% in the elderly women. From the mineral and water fractions in the fat-free mass the true density of the fat-free mass was calculated as 1·1070 (SD 0·0047) kg/l in the young females and 1·0970 (SD 0·0088) kg/l in the elderly women (P<0·001). This study shows that the single methods have considerable mean and individual biases compared with the four-compartment model, but that a three-compartment model calculated from density and total body water offers an acceptable alternative. The difference in calculated density of the fat-free mass between the young and the elderly women shows the need to adapt Siri's formula for specific groups.