The components of reliability for eight anthropometric measures were studied in 95 male and 134 female subjects from the Second National Health and Nutrition Examination Survey (NHANES II). The contributions to unreliability variance (Sr2) that occur as a result of measuring errors (Sp2, imprecision variance) and of intrasubject fluctuations in a measurement due to physiologic factors (Sd2, undependability) were estimated (Sr2 = Sp2 + Sd2). Unreliability was then related to the between-subject variance (S2) to estimate the reliability (R = 1 - (Sr2/S2)) of the measurement. Four of the anthropometric measurements (weight, height, sitting height, and arm circumference) had reliabilities in excess of R = 0.97. In the first three of these, measurement imprecision made up two thirds or less of unreliability, and undependability (Sd2) was stable by two weeks. Lesser but still acceptable reliabilities were obtained for triceps and subscapular skinfolds, bitrochanteric breadth, and elbow breadth (R = 0.81-0.95). For these variables imprecision (Sp2) was the major source of error. Furthermore, the unreliability (Sr2) between observers was twice as high or more than the unreliability within observers for these variables, evidence that imprecision (Sp2) is the single most important source of unreliability in these anthropometric measurements. Unreliability standard deviations of skinfolds increased in a linear manner with skinfold thickness corresponding to an unreliability coefficient of variation of 13-19 per cent. None of the other measurements showed such scale effects. Analyses of the kind suggested will help epidemiologists decide whether reliability can be increased by improving precision, and whether there is a need to improve reliability in the first place. Reliability appears to be adequate for all anthropometry in the NHANES II.