Interobserver variations for measurements of the Cobb angle on radiographs of patients who had kyphosis were comparable with those on the radiographs of patients who had scoliosis. Four staff orthopaedists and one physical therapist measured eight radiographs that showed scoliosis and twenty that showed kyphosis. The measurements were made on two occasions and in random order. For scoliosis, the average difference between readings was 3.8 degrees, and 95 per cent of the differences were 8 degrees or less (range, 0 to 10 degrees). These findings were in keeping with those of other published reports. For kyphosis, the average difference between readings was 3.3 degrees, and 95 per cent of the differences were 7 degrees or less (range, 0 to 30 degrees). One investigator rated the kyphosis radiographs with respect to clarity. There was a trend to less variation with clearer radiographs, but this was not significant. The end-vertebrae were pre-selected for some radiographs and were freely chosen by the interpreter for others. Reliability was not significantly improved when the end-vertebrae of the curve had been pre-selected. Using the statistical method called tolerance limits, we determined that if one were to be 95 per cent confident that a measured difference represented a true change, the difference would have to be 10 degrees for scoliosis radiographs and 11 degrees for kyphosis radiographs. The probability that a measured difference is due to measurement error alone (that is, a false-positive reading) was calculated.