To determine whether, with improvements in radioimmunoassay techniques, duplication is still necessary, the differences between duplicate results for a range of assays done routinely over one month were examined retrospectively. Differences over 10% between duplicates were found in 104/779 (13%) of assays for thyroid stimulating hormone, 27/180 (15%) for total thyroxine, 44/378 (12%) for cortisol, 15/355 (4%) for follicular stimulating hormone, 20/356 (6%) for luteinising hormone, and none for alpha fetoprotein (0/256). In only two of 779 patients (0.26%) would the different result of a pair of thyroid stimulating hormone duplicates have led to different courses of action by the laboratory. None of the other differences in any assay would have resulted in a potential misclassification. Although replication of assays will give more correct results by pure scientific criteria, the improvement is rarely clinically important and the financial cost is considerable.