52 patients with refractory or relapsed acute myeloid leukaemia (AML) were randomly assigned to receive a combination of high-dose cytosine arabinoside (HD Ara-C), 3 g/m2/d and either mitoxantrone (MTX), 7 mg/m2/d (5 mg if older than 60 yr) or m-amsacrine (AMSA), 120 mg/m2/d (90 mg if older than 60 yr) for 5 d. The overall response rate was 50% and did not differ significantly in the two groups (46% for AMSA and 56% for MTX, p = 0.415). The median survival was 11 months (8 months for AMSA and 12 months for MTX, p = 0.326) and the median duration of complete remission (CR) was 11 months for AMSA and 12 months for MTX (p = 0.643). In relapsed patients, the only significant predictive factor for obtaining a complete response was the length of first complete remission. Patients with a first CR shorter than 6 months had a CR rate of 36% while it was 65% if the first CR lasted more than 6 months (p = 0.03). Severe (WHO grade III-IV) gastro-intestinal toxicity was more frequent in the AMSA group (27% vs 4%, p = 0.021). Treatment-related death occurred in 4 patients in the AMSA group and in 2 patients in the MTX group (p = 0.097). We conclude that neither of these two treatment modalities was shown to be superior in terms of CR rate and survival, with a better tolerance for MTX.