Risk stratification for acute myeloid leukemia (AML) uses molecular and cytogenetic abnormalities identified at diagnosis. Response to therapy informs risk, and morphology continues to be used more frequently than flow cytometry. Herein, the largest cohort of pediatric patients prospectively assessed for measurable residual disease (MRD) by flow cytometry (N = 784) is reported. The "difference from normal" (ΔN) technique was applied: 31% of all patients tested positive (AML range, 0.02% to 91%) after the first course of treatment on Children's Oncology Group study AAML0531. Detection of MRD following initial chemotherapy proved the strongest predicator of overall survival (OS) in univariable and multivariable analyses, and was predictive of relapse risk, disease-free survival, and treatment-related mortality. Clearance of MRD after a second round of chemotherapy did not improve survival. The morphologic definition of persistent disease (>15% AML) failed 27% of the time; those identified as MRD- had superior outcomes. Similarly, for patients not achieving morphologic remission (>5% blasts), 36% of patients were MRD- and had favorable outcomes compared with those who were MRD+ (P < .001); hence an increase in myeloid progenitor cells can be favorable when ΔN classifies them as phenotypically normal. Furthermore, ΔN reclassified 20% of patients in morphologic remission as having detectable MRD with comparable poor outcomes. Retrospective analysis using the relapse phenotype as a template demonstrated that 96% of MRD- patients had <0.02% of the relapse immunophenotype in their end of induction 1 marrow. Thus, the detection of abnormal myeloid progenitor cells by ΔN is both specific and sensitive, with a high predictive signal identifiable early in treatment. This trial was registered at www.clinicaltrials.gov as #NCT00372593.