The Childrens Cancer Study Group (CCSG) CCG-160 protocol series was designed to evaluate prognostic factors in acute lymphoblastic leukemia (ALL). Patients were assigned to one of three prognostic groups based upon initial WBC count and age. To determine the optimal duration of therapy, CCG-160 patients completing 2 years of treatment in continuous remission were randomized ("late randomization") to discontinue therapy or receive another year of maintenance therapy. The prognostic significance of early response to induction therapy, as measured by the percentage of lymphoblasts in the day-14 bone marrow (d14 BM) aspirate, was evaluated in 2,516 children. For 1,490 patients with complete data, the status of the d 14 BM was a highly significant predictor of disease-free survival (DFS) by univariate and multivariate analysis (P less than .0001). The observed/expected (O/E) failure rate in patients with d14 M1 (less than 5% blasts), M2 (4% to 25% blasts), or M3 (greater than 25% blasts) BM rating who were subsequently M1 on day 28 or day 42, was .87, 1.59, and 2.30, respectively (P less than .0001). Patients with M2 or M3 d14 BM were more likely to have L2 ALL (modified French-American-British [FAB] morphologic classification), P less than .001. The significance of the d14 BM rating persisted after correction was made for WBC count and clinical prognostic groups using current CCSG criteria, except in infants less than 12 months of age. The d14 BM was also the most significant predictor of DFS in 975 patients after late randomization at 2 years following diagnosis. The O/E failure rate in patients with d14 M1, M2, or M3 BM was .88, 1.78, and 2.02, respectively (P = .0002, trend). Other significant predictors of late relapse were prognostic groups (P = .0003, trend) and initial WBC count (P = .004, trend). Predictive for both early and late relapse of ALL, early response should be monitored closely and alternative treatment regimens should be considered for slow responders.