Diaziquone given as a bolus has not been effective in patients with relapsed or refractory leukemia. Because of in vitro data suggesting enhancement of diaziquone-induced cytotoxicity for human and murine leukemia cells with increased duration of drug exposure and the relatively short terminal plasma half-life of diaziquone, 49 patients (34 acute nonlymphocytic leukemia [ANLL], six chronic myelogenous leukemia in blast crisis [CML-B], five acute lymphocytic leukemia [ALL], four 2 degrees ANLL) with leukemia were given diaziquone as a continuous infusion for seven days. The maximum tolerated dose was 28 mg/m2/d for seven days. The dose-limiting toxicity was the duration of bone marrow aplasia (median, 49 days to greater than 500 PMNs in responders; range, 28 to 101 days). Nonhematologic toxicity was minimal. Responses occurred only in patients with relapsed ANLL, of whom 26 were treated at effective doses. There were six complete responses (CR) (23%) and two partial responses (PR) (8%), although five of eight responders never achieved platelet counts greater than 100,000/microL. Thrombocytopenia in these patients was felt to be a manifestation of diaziquone effect, not persistence of leukemia. The median duration of CR was 195 days (range, 88 to 860+). One patient had active CNS leukemia at the start of treatment and has had a durable (28+ month) CR in both sites of disease. Diaziquone produced prolonged aplasia in patients with secondary ANLL and CML-B (five of ten patients died aplastic), whereas patients with ALL all had regrowth of leukemia and two failed to become aplastic. The lack of significant nonhematologic toxicity and the activity in patients with relapsed ANLL render diaziquone of interest as second-line therapy or consolidation therapy in first remission for patients with ANLL.