Leukemia cell karyotypes were determined at diagnosis for 136 of 159 consecutive patients with acute lymphoblastic leukemia (ALL) who were followed for up to 35 mo. Ninety patients (67%) had abnormal karyotypes. Five chromosome categories were designated, based on the distribution of modal numbers: hyperdiploid greater than 50 (n = 41), hyperdiploid 47–50 (n = 18), pseudodiploid (n = 28), normal (n = 46), and hypodiploid (n = 3). Treatment response was assessed for the categories in terms of time to failure (induction failure, first relapse, or death). Children in the hyperdiploid greater than 50 category had the best responses to treatment, with only 2 failures, and those in the pseudodiploid category had the poorest (p less than 0.001). The remaining 3 chromosome categories had intermediate responses and formed a third prognostic group. This same influence of chromosome number on time to failure was evident within the 2 clinical prognostic groups: high risk, signified by a leukocyte count greater than 100 X 10(9)/liter, meningeal leukemia, mediastinal mass, or the presence of blasts that formed rosettes with sheep erythrocytes at 37 degrees C, and standard risk, indicated by the absence of these features. The influence of chromosome number on time to failure was also the same within the historically favorable prognostic group that had common ALL. Results of a multivariate analysis indicated that chromosome number was the strongest single predictor of outcome (p less than 0.001) and was the only variable that added significant prognostic information to leukocyte count (p less than 0.001). The combination of chromosome number and leukocyte count should more clearly distinguish patients with ALL at low or high risk of relapse.