Acute myeloid leukemia (AML) is a heterogeneous disease with respect to biology and clinical course. Until now the basis for prognostic evaluation and therapeutic decision has been the karyotype, genetic FLT3 abnormalities and the initial chemotherapy response. A question that has emerged is if extensive gene expression analysis may supplement or partly replace current diagnostics. In an attempt to address this question, we performed cDNA microarray analysis on peripheral blood samples of 25 patients with newly diagnosed AML with high blast counts. The patients were randomly selected from a large group of consecutive patients. Leave-one-out crossvalidation (LOOCV) showed with high accuracy that gene expression classifiers could predict if leukaemia samples belonged to the FAB AML-M1 or to the FAB AML-M2 groups. An unsupervised two-dimensional hierarchical cluster analysis generated 3 patient subgroups. Except for an accumulation of samples classified as FAB M1 and M2 in cluster 3, there was no evident relationship between the clusters and the FAB classification. Each subgroup displayed clearly distinguished gene expression patterns validated using real-time quantitative PCR analysis. The identification of specific gene expressions that together constitute regulatory modules must complement cluster analyses in order to achieve an accurate basis for prognosis and prediction.