A total of 16 obese women (body mass index (BMI) 30-43 kg m(-2)) participated in a weight reduction study. Before and after a weight loss of 11.7 +/- 7.4 kg (mean +/- SD), body composition was assessed by dual energy X-ray absorptiometry (DXA), and total body potassium counting (TBK). These measurements were compared with bioimpedance analysis (BIA) by applying 11 predictive BIA equations published in the literature. Predictive equations for the present study population were developed, with the use of fat-free mass (FFM) as assessed by TBK and DXA as references in multiple regression analysis. The results of the BIA equations varied widely; FFM was generally overestimated by BIA as compared with DXA and TBK before and after weight loss. During weight loss, the FFM did not change, as estimated by DXA (1.3 +/- 2.3 kg, p > 0.05) and TBK (0.9 +/- 2.9 kg, p > 0.05). The recorded change in impedance (R) was also insignificant. Three BIA equations from the literature, which were not specific for the degree of obesity in the present study group, predicted changes in FFM (from 0.5 + 3.6 to 2.4 +/- 4.4kg, p > 0.05) that were comparable with those estimated by the reference methods. Eight equations from the literature, which included equations specific for the degree of obesity in the study group, and the group specific equations developed for the present population predicted significant changes in FFM during weight loss (from 2.3 +/- 3.0 to 5.0 +/- 3.0 kg, p < 0.05). We conclude that in obesity most predictive equations are unable to predict static body composition and are not reproducible for individuals over time. However, a significant or insignificant change in R (without accompanying predictive equations) may be used to indicate whether FFM is lost or preserved in groups of obese subjects.