There is renewed interest in Siri's classic three-compartment (3C) body composition model, requiring body volume (BV) and total body water (TBW) estimates, because dual-energy X-ray absorptiometry (DEXA) and in vivo neutron activation (IVNA) systems cannot accommodate subjects with severe obesity. However, the 3C model assumption of a constant ratio (alpha) of mineral (M) to total body protein (TBPro) and related residual mass density (D(RES)) based on cadaver analyses might not be valid across groups differing in sex, race, age, and weight. The aim of this study was to derive new 3C model coefficients in vivo and to compare these estimates to those derived by Siri. Healthy adults (n = 323) were evaluated with IVNA and DEXA and the measured components used to derive alpha and D(RES). For all subjects combined, values of alpha and D(RES) (means +/- SD, 0.351 +/- 0.043; 1.565 +/- 0.023 kg/l) were similar to Siri's proposed values of 0.35 and 1.565 kg/l, respectively. However, alpha and D(RES) varied significantly as a function of sex, race, weight, and age. Expected errors in percent body fat arising by application of Siri's model were illustrated in a second group of 264 adults, including some whose size exceeded DEXA limits but whose BV and TBW had been measured by hydrodensitometry and (2)H(2)O dilution, respectively. Extrapolation of predictions by newly developed models to very high weights allows percent fat error estimation when Siri's model is applied in morbidly obese subjects. The present study results provide a critical evaluation of potential errors in the classic 3C model and present new formulas for use in selected populations.