The inter-observer variation in simple bedside assessments of body composition has been evaluated in 12 healthy adult subjects (6 male and 6 female) by 6 observers. Systematic bias between observers was found to be evident in each of the basic measurements (except height) and in each of the estimates of body composition (except those derived from weight and height alone). The largest residual coefficients of variation (rCV) for the basic measurements were found for skinfold thickness (11-18 per cent for individual skinfold thicknesses, and 9 per cent for the sum of four skinfold thicknesses), and the lowest for weight (0.01 per cent for digital scales, and 0.05 per cent for beam balance) and height (0.4 per cent). The rCV for whole-body resistance (1.2 per cent), forearm resistance (5.4 per cent) and near infra-red interactance (optical density 1: 5.6 per cent and optical density 2: 6.2 per cent) measurements were found to have intermediate values. The variability in the estimate of body fat obtained by skinfold thickness (rCV = 4.6 per cent) and near infra-red interactance methods (rCV = 4.2 per cent) was found to be greater than that from the resistance method (rCV = 2.6 per cent) and by methods based on weight and height alone (rCV = 1.1 per cent). The variability in the estimate of fat-free mass showed the same trend as that for body fat, but the rCVs were less due to the greater mass of the fat-free body compared to body fat.(ABSTRACT TRUNCATED AT 250 WORDS)