Accurate methods for determining body fat mass during reproduction are necessary to evaluate energy balance. However, determination of fat mass is complicated during pregnancy by the accretion of water, which invalidates assumptions underlying standard two-compartment models. The extent to which the variability in body water during pregnancy invalidates use of pregnancy-corrected two-compartment models for determination of fat mass in individual women is unknown. Moreover, it is unclear whether body water returns to nonpregnant values by 2 wk postpartum, which is frequently used as the baseline in studies of postpartum women. The present study uses a four-component model as a criterion for evaluating two- and three-component models. Fifty-six healthy, normotensive women between the ages of 19 and 35 y were studied at 36 +/- 1 wk gestation and 15 +/- 2 d postpartum. Total body water (TBW), total body potassium (TBK), body density, and bone mineral content were measured by deuterium dilution, whole-body potassium counting, hydrodensitometry, and dual-energy X-ray absorptiometry (postpartum only), respectively. At 2 wk postpartum, hydration and density of fat-free mass (FFM) had not returned to nonpregnant values, and differed between lactating and nonlactating women (P < 0.05). Accordingly, standard TBW and body density estimates of fat mass differed from four-component estimates at both time points (P < 0.005). Moreover, our data indicate that even when pregnancy-specific values for hydration or density of FFM are used in TBW and body density models, individual fat mass estimates may differ by > 3 kg from the four-component value. Fat mass by TBK may differ by > 10 kg from fat mass by the four-component model during pregnancy, and by 6 kg postpartum. Use of standard two-compartment models to estimate fat mass results in significant error both during pregnancy and at 2 wk postpartum. Pregnancy-corrected two-compartment models produce reliable mean fat mass estimates during pregnancy, but individual fat mass estimates may vary widely from four-component values.