Gene expression analyses were performed on 121 consecutive childhood leukemias (87 B-lineage acute lymphoblastic leukemias (ALLs), 11 T-cell ALLs and 23 acute myeloid leukemias (AMLs)), investigated during an 8-year period at a single center. The supervised learning algorithm k-nearest neighbor was utilized to build gene expression predictors that could classify the ALLs/AMLs according to clinically important subtypes with high accuracy. Validation experiments in an independent data set verified the high prediction accuracies of our classifiers. B-lineage ALLs with uncharacteristic cytogenetic aberrations or with a normal karyotype displayed heterogeneous gene expression profiles, resulting in low prediction accuracies. Minimal residual disease status (MRD) in T-cell ALLs with a high (>0.1%) MRD at day 29 could be classified with 100% accuracy already at the time of diagnosis. In pediatric leukemias with uncharacteristic cytogenetic aberrations or with a normal karyotype, unsupervised analysis identified two novel subgroups: one consisting mainly of cases remaining in complete remission (CR) and one containing a few patients in CR and all but one of the patients who relapsed. This study of a consecutive series of childhood leukemias confirms and extends further previous reports demonstrating that global gene expression profiling provides a valuable tool for genetic and clinical classification of childhood leukemias.