The present randomized trial addressed the pending question whether cytosine arabinoside (AraC) should be given at high or intermediate dose to patients with relapsed or refractory acute myeloid leukemia. Based upon the previously established regimen of the sequential administration of AraC and mitoxantrone (S-HAM) patients below 60 years of age were randomized to receive AraC at either 3.0 g/m2 vs. 1.0 g/m2 per dose while older patients were randomly assigned to either 1.0 g/m2 or 0.5 g/m2 AraC. Concurrent pharmacokinetic analyses were performed to determine the plasma AraC pharmacokinetics as well as the intracellular AraCTP peak concentrations and retention times. At the present stage 65 patients are evaluable for response and toxicity. Complete remissions were achieved at similar frequencies for patients treated with 3.0 g/m2 or 1.0 g/m2 AraC with 56% and 50%, respectively. Reasons for failure were different, however, with a higher incidence of resistant disease in patients treated with 1.0 g/m2 AraC and more early deaths in the higher dose treatment group. Pharmacokinetic studies indicated a homogeneous distribution of AraC plasma concentrations but a substantial interpatient variability for intracellular AraCTP peak concentrations and retention times.