Seven severely obese, outpatient dieters lost weight (mean ± SEM, 14 ± 1 kg), and the composition of weight lost was determined by six different models. Total body water (TBW), total body potassium (TBK), and body density, bone mineral content, and fat as determined by dual photon absorptiometry (DPA) were measured while subjects were weight-stable, before and after weight loss. Fat loss was calculated by three two-compartment models (2C-TBW, 2C-TBK, and hydrodensitometry [2C-HD]), one three-compartment model (HD with correction for water content of fat-free mass (FFM], 3C), and one four-compartment model (HD with correction for water and mineral content of FFM, 4C), and was measured directly by DPA. Mean composition of weight loss was similar for all models (mean weight lost as fat: 89% for DPA, 91.5% for 4C, 89% for 3C, 88.6% for 2C-HD, and 87% for 2C-TBW) except 2C-TBK (weight lost as fat, 66%). There was a much wider range of individual values for the 2C-TBW and 2C-TBK models (17% to 138% and 18% to 93%, respectively) than for the multicompartment models (63% to 112%) and DPA (76% to 107%). Almost opposite results were obtained for the same individual when using the 2C-TBK and 2C-TBW models. The discrepancy between these models was due to the inverse relationship between changes in TBW and TBK in the group as a whole (r = - .34, NS). In addition, TBK loss was found to be dependent on the initial level of hyperinsulinemia, calculated as the area under the 2-hour oral glucose tolerance curve. In conclusion, in severely obese patients, fat-loss measurements by presently available methods are not very consistent for individuals, although group averages are similar. Individual differences are due to variable changes in body fluid and potassium compartments, and multicompartment models are recommended to correct for this.