Using percentage body fat (BF%) from a three-compartment (3C) model (body density from underwater weighing (UWW) and bone-mineral mass from dual-energy X-ray absorptiometry (DXA)) as a criterion, we studied the accuracy of UWW, DXA, two skinfold equations, and two bioimpedance (BIA) equations. Thirty-four women (aged 16–20 years) with BF% 13·5–31·1 volunteered. UWW underestimated BF% by −0·5 BF% (95 % CI: −1·0; −0·02), whereas DXA overestimated it by 7·3 BF% (95% CI5.8;8·8). Skinfolds underestimated and BIA overpredicted BF%. The differences between 3C and UWW, skinfolds (Durnin & Womersley, 1974) and BIA (Deurenberg et al. 1990) were dependent (range of r values: −0.63 to −0.79; P < 0·0001) on BF%, causing an overestimation of lean subjects' (UWW, BIA) or an underestimation of normal-weight subjects' (UWW, skinfolds) BF%. The 3C model and UWW gave comparable body-composition results for healthy young women with BF% of approximately 20–25. Based on a significant mean difference from the 3C model, and a large standard error of the estimate, we do not regard DXA as superior to skinfolds or BIA to assess BF%.