Many patients with acute myeloid leukemia (AML) are incurable with chemotherapy and may benefit from novel approaches. One such approach involves the transfer of T cells engineered to express chimeric antigen receptors (CARs) for a specific cell-surface antigen. This strategy depends upon preferential expression of the target on tumor cells. To date, the lack of AML-specific surface markers has impeded development of such CAR-based approaches. CD123, the transmembrane α chain of the interleukin-3 receptor, is expressed in the majority of AML cells but is also expressed in many normal hematopoietic cells. Here, we show that CD123 is a good target for AML-directed CAR therapy, because its expression increases over time in vivo even in initially CD123^dim populations, and that human CD123-redirected T cells (CART123) eradicate primary AML in immunodeficient mice. CART123 also eradicated normal human myelopoiesis, a surprising finding because anti-CD123 antibody-based strategies have been reportedly well tolerated. Because AML is likely preceded by clonal evolution in “preleukemic” hematopoietic stem cells, our observations support CART123 as a viable AML therapy, suggest that CART123-based myeloablation may be used as a novel conditioning regimen for hematopoietic cell transplantation, and raise concerns for the use of CART123 without such a rescue strategy.