Approximately 40% of children with acute myeloid leukemia (AML) who respond to initial therapy subsequently relapse. Multidimensional flow cytometry employing a standardized panel of monoclonal antibodies enables the detection of small numbers of occult leukemic cells that persist during therapy using technology adaptable by most clinical laboratories. We performed a prospective, blinded evaluation of bone marrow specimens obtained from 252 pediatric patients with de novo AML to determine whether detection of occult leukemia defined as more than or equal to 0.5% blasts with aberrant surface antigen expression as determined by flow cytometry was predictive of subsequent relapse. Occult leukemia was detected in 41 (16%) of the 252 patients who responded to initial induction therapy. In time-dependent multivariate analyses that controlled for allogeneic marrow transplantation, variable intervals between sample submission, age, sex, white blood cell count at diagnosis, presence of splenomegaly or hepatomegaly, and presence of more than 15% blasts in the marrow after the first course of induction, patients harboring occult leukemia were 4.8 times more likely to relapse (95% confidence interval CI = 2.8 to 8.4,P < .0001) and 3.1 times more likely to die (95% CI; 1.9 to 5.1, P < .0001) than those lacking leukemia detectable by flow cytometry. In this analysis, flow cytometric evidence of leukemia after the initiation of therapy emerged as the most powerful independent prognostic factor associated with poor outcome. Among patients in whom a marrow sample was available for analysis at the end of consolidation therapy, overall survival at 3 years was 41% versus 69% for patients with and without occult leukemia, respectively (P = .0058).