Seventy-two adults were treated for acute myelogenous leukaemia (AML). Forty-two had previously untreated AML and 30 had AML after a preleukaemic phase, refractory AML or relapsed AML. The previously untreated patients received a 7-day course of cytosine arabinoside (100 or 200 mg/m2 daily), daunorubicin and vincristine while the remaining patients received a 7-day course of cytosine-arabinoside (1 g/m2 q 12h for 6 days) and amsacrine (on day 7). The percentage of malignant cells and the reduction in the percentage of malignant cells were determined by means of bone marrow aspirates taken on day 6 of the chemotherapy course and at the time of diagnosis. Both variables correlated significantly with the ultimate treatment outcome; the reduction in the percentage of malignant cells correlated even more significantly than the absolute percentage malignant cells in the day-6 bone marrow. By means of multiple regression analysis it became possible to calculate the probability of achieving complete remission for the individual patient; this is given by the equation: probability = 1.9-0.009X (% malignant cell reduction). In addition, the mean percentage of malignant cells in the day-6 bone marrow was significantly higher for patients who failed to achieve than those who entered complete remission. Eighty-six per cent of the patients with less than 20% malignant cells on day 6 entered remission, while 75% of the patients with more than 21% malignant cells failed to achieve complete remission (p less than 0.001). Although all of these calculations support the predictive value of the day-6 bone marrow aspirate, the 95% confidence intervals are too large to allow reliable and safe predictions; therefore more patients must be studied to demonstrate the reliability of this test.