1. Body composition was assessed in 28 healthy subjects (body mass index 20-28 kg/m2) by dual-energy X-ray absorptiometry, deuterium dilution, densitometry, 40K counting and four prediction methods (skinfold thickness, bioelectrical impedance, near-i.r. interactance and body mass index). Three- and four-component models of body composition were constructed from combinations of the reference methods. The results of all methods were compared. Precision was evaluated by analysis of propagation of errors. The density and hydration fraction of the fat-free mass were determined. 2. From the precision of the basic measurements, the propagation of errors for the estimation of fat (+/- SD) by the four-component model was found to be +/- 0.54 kg, by the three-component model, +/- 0.49 kg, by deuterium dilution, +/- 0.62 kg, and by densitometry, +/- 0.78 kg. Precision for the measurement of the density and hydration fraction of fat-free mass was +/- 0.0020 kg/l and +/- 0.0066, respectively. 3. The agreement between reference methods was generally better than between reference and alternative methods. Dual-energy X-ray absorptiometry predicted three- and four-component model body composition slightly less well than densitometry or deuterium dilution (both of which greatly influence these multi-component models). 4. The hydration fraction of fat-free mass was calculated to be 0.7382 +/- 0.0213 (range 0.6941-0.7837) and the density of fat-free mass was 1.1015 +/- 0.0073 kg/l (range 1.0795-1.1110 kg/l), with no significant difference between men and women for either. 5. The results suggest that the three- and four-component models are not compromised by errors arising from individual techniques.(ABSTRACT TRUNCATED AT 250 WORDS)