We found in previous studies that the cytotoxicity of 4-hydroperoxycyclophosphamide (4-HC) toward normal and malignant cells in autologous bone marrow grafts was modulated by the variable erythrocyte content of the incubation mixture. This affected both the kinetics of engraftment and the probability of leukemic relapse. To decrease this variability in 4-HC cytotoxicity, we initiated a trial of density-gradient separation of the autografts to remove erythrocytes and other mature blood cells before the 4-HC treatment. A series of 64 consecutive patients was studied. Light-density (sp. gr. much less than 1.077 g/ml) cells were isolated in a 2-step procedure by Ficoll-diatrizoate separation of previously isolated buffy-coat cells. Overall cell recovery after buffy-coat and gradient separations was 28.9 +/- 11.7% ( +/- SD). These cells were then treated with 60 micrograms/ml of 4-HC. The erythrocyte content of the incubation mixtures averaged 0.8%, and did not predict percent CFU-GM survival after the 4-HC incubation. The variability in CFU-GM survival after the purge was less than for previous patients who received buffy-coat cells (with erythrocyte contents ranging from 4-22%) treated with 100 micrograms/ml of 4-HC, indicating that a more uniform purge was achieved. Although the selected dose of 4-HC (60 micrograms/ml) yielded a lower CFU-GM survival, the kinetics of engraftment did not significantly differ between the recipients of light-density or buffy-coat cells. No patient required use of a 'backup' graft because of engraftment failure.