Between June 1981 and April 1986 63 patients with acute myeloid leukaemia (AML) in first remission and HLA-identical sibling donors were entered into a prospective study comparing cyclophosphamide (CY) + total body irradiation (TBI) with melphalan + TBI as conditioning therapy prior to transplantation. Thirty-six patients received CY/TBI and 27 received melphalan/TBI. The actuarial probability of remaining in remission for patients receiving melphalan/TBI was 94% compared with 66% following CY/TBI (p greater than 0.01). By including a comparable group of 41 patients with AML in first remission, conditioned with CY/TBI prior to the onset of the study, the greater anti-leukaemic effect of melphalan/TBI compared to CY/TBI was unchanged and statistically the chance of this being wrong is 1 in 10 (p less than 0.1). The overall survival in remission of both arms of the study was the same with 15/27 patients (55%) surviving in remission following melphalan/TBI compared with 19/36 patients (53%) following CY/TBI. The benefit obtained in reduced relapse was offset by the combined nephrotoxic effect of melphalan and cyclosporin which was not identified until the programme had been underway for a period of time. This shows that misinterpretation of 'no survival advantage' for the new treatment may occur due to unforeseen and preventable toxicities.