One hundred four children with advanced leukemia in relapse (74 with acute lymphocytic leukemia [ALL] and 30 with acute nonlymphocytic leukemia [ANLL]) received AMSA (4'-(9-Acridinylamino) methanesulfon -m-anisidide) at a dose of 120 mg/m2/day for 5 days (Regimen I) or 60 mg/m2/day for 10 days (Regimen II). Children with ALL were randomized between Regimens I and II (31 and 36 evaluable patients, respectively). All 29 evaluable patients with ANLL were treated on Regimen I. Eighty-eight percent of evaluable patients experienced severe or life-threatening toxicity, with no statistical differences between Regimens I and II. Bacterial or fungal infections (considered life-threatening or fatal) occurred in 17 children with ALL and in 7 with ANLL. Fatal cardiac toxicity occurred in one patient. Complete or partial response occurred in 25.0% (SE = 8.8%), 28.1% (SE = 8.0%), and 25.9% (SE = 8.4%) of evaluable patients on ALL Regimen I, ALL Regimen II, and ANLL, respectively. However, responses were of short duration (16-91 days). There was no significant difference in the duration of survival from treatment start for the two ALL regimens (P = 0.46). The median duration of survival for ANLL patients was significantly longer (P = 0.004) than that of ALL patients treated on Regimens I and II combined. Eighty-two percent of the complete or partial responses (18 of 22) occurred after the first course of AMSA. At the dose schedules investigated, and in a heavily pretreated patient population, AMSA had activity in childhood leukemia. However, the high incidence of severe, life-threatening, or fatal infections meant that the quality and quantity of responses and survival was not commensurate with the toxicity, and that it would be difficult to incorporate this drug into combination chemotherapy with other myelosuppressive agents.