Percent body fat (%BF) was assessed in 171 men with underwater weighing (UWW), seven-site skinfolds (7 SF), and near-infrared spectrophotometry (NIR). NIR was determined at eight sites, including biceps, triceps, axilla, chest, abdomen, suprailium, subscapula and thigh. Analysis of variance and Dunnett's post-hoc procedure revealed that NIR significantly (P less than 0.05) underpredicted %BF when using the biceps (12.9%), chest (11.3%), abdomen (10.2%), subscapula (11.3%) and thigh (9.9%) sites compared to the criterion %BF measured with UWW (13.4% +/- SD = 6.49). %BF with 7 SF was only 0.3% higher than %BF with UWW (r = 0.94, standard error of estimate = 2.9%). Correlation coefficients between SF thickness and NIR optical density readings at 940 nm (OD1) and 950 nm (OD2) wavelengths ranged from r = -0.30 (subscapula) to r = -0.67 (biceps) for OD1 and r = -0.39 (axilla) to r = -0.68 (biceps) for OD2. Multiple linear regression showed that the OD readings were not significant predictors of UWW %BF when using body mass, stature, activity level, and frame size as predictors, except for the OD2 reading at the biceps site. Stepwise regression revealed that NIR OD readings did not predict additional variance in %BF beyond mass, stature, activity level, frame size, or 7 SF. It was concluded that 7 SF are better predictors of %BF than NIR, and that NIR at multiple sites is inconsistent in assessing body composition.