The accuracy of a variety of in vivo body-composition techniques (densitometry, total body water, skinfold thicknesses, whole-body impedance and resistance, body mass index, and two three-compartment models) was assessed by comparison with fat balance. Three subjects were overfed and three underfed while confined to a 30-m3 whole-body calorimeter continuously for 12 d. Mean weight changes were +2.90 kg during overfeeding and -3.47 kg during underfeeding. The change in fat mass accounted for 37.1% during overfeeding and 59.3% during underfeeding. In comparison with energy and nitrogen balance, a three-compartment model yielded the least bias and greatest precision. The smallest change in fat mass that can be measured by such a method in an individual subject is 1.54 kg (2 SD). Of the prediction techniques considered, skinfold thicknesses or the body-mass-index formula appear to be more precise than estimates based on resistance or impedance.