Bioelectrical impedance analysis (BIA) is an affordable, non-invasive, easy-to-operate, and fast alternative method to assess body composition. However, BIA tends to overestimate the percent body fat (%BF) in lean elderly and underestimate %BF in obese elderly people. This study examined whether proximal electrode placement eliminates this problem. Forty-two elderly men and women (64-96 years) who had a wide range of BMI [22.4 +/- 3.3 kg/m(2) (mean +/- SD), range 16.8-33.9 kg/m(2)] and %BF (11.3-44.8%) participated in this study. Using (2)H and (18)O dilutions as the criterion for measuring total body water (TBW), we compared various BIA electrode placements; wrist-to-ankle, arm-to-arm, leg-to-leg, elbow-to-knee, five- and nine-segment models, and the combination of distal (wrists or ankles) and proximal (elbows or knees) electrodes. TBW was most strongly correlated with the square height divided by the impedance between the knees and elbows (H(2)/Z (proximal); r = 0.965, P < 0.001). In the wrist-to-ankle, arm-to-arm, leg-to-leg, and five-segment models, we observed systematic errors associated with %BF (P < 0.05). After including the impedance ratio of the proximal to distal segments (P/D) as an independent variable, none of the BIA methods examined showed any systematic bias against %BF. In addition, all methods were able to estimate TBW more accurately (e.g., in the wrist-to-ankle model, from R(2) = 0.90, SEE = 1.69 kg to R(2) = 0.94, SEE = 1.30 kg). The results suggest that BIA using distal electrodes alone tends to overestimate TBW in obese and underestimate TBW in lean subjects, while proximal electrodes improve the accuracy of body composition measurements.