In a prospective trial in 284 children with B-lineage acute lymphoblastic leukemia (ALL), we assessed the clinical utility of real-time quantitative polymerase chain reaction analysis of antigen receptor gene rearrangements for detection of minimal residual disease (MRD) to identify children at high risk of relapse. At the end of induction therapy, the 5-year risk of relapse was 5% in 176 children with no detectable MRD and 44% in 108 children with detectable MRD (P < .001), with a linear association of the level of MRD and subsequent relapse. Recursive partitioning and clinical characteristics identified that the optimal cutoff level of MRD to predict outcome was 10(-3). The 5-year risk of relapse was 12% for children with MRD less than one leukemia cell per 10(3) normal cells (low MRD) but 72% for children with MRD levels greater than this level (high MRD) (P < .001) and children with high MRD had a 10.5-fold greater risk of relapse. Based upon these results we have altered our treatment regimen for children with B-lineage ALL and children with MRD levels greater than or equal to 10(-3) at the end of 4 weeks of multiagent induction chemotherapy now receive intensified treatment to attempt to decrease their risk of subsequent relapse.