In animal models of various cancer types, so-called cancer stem cells (CSCs) have been shown to be important for disease maintenance and therapy resistance. However, direct relevance of these findings for human tumors has yet to be demonstrated. In the case of acute myeloid leukemia (AML), Eppert et al. now show that characteristic gene expression profiles of functionally defined CSC populations prognosticate treatment outcome in patients. To obtain CSC expression profiles of human patients, the authors sorted 16 patient samples into four cell populations based on the expression of surface markers, and assayed the ability of each subpopulation to engraft in highly immunodeficient mice. They analyze the gene expression profile of CSC-containing fractions to define a CSC-specific gene signature, which they use to classify AML patients into high-risk and low-risk groups. The high-risk group has an up to 2.4 times higher probability to succumb to the disease. Interestingly, gene signatures derived from cell populations defined on the basis of cell surface markers only (without functional validation) cannot distinguish low-risk from high-risk patients. (Nat. Med. 17, 1086–1093, 2011)