A fundamental difficulty in testing "targeted therapies" in acute myeloid leukemia (AML) is the limitations of preclinical models in capturing inter-and intrapatient genomic heterogeneity. Clinical trials typically focus on single agents despite the routine emergence of resistant subclones and experience in blast-phase chronic myeloid leukemia and acute promyelocytic leukemia arguing against this strategy. Inclusion of only relapsed-refractory, or unfit newly diagnosed, patients risks falsely negative results. There is uncertainty as to whether eligibility should require demonstration of the putative target and regarding therapeutic end points. Although use of in vivo preclinical models employing primary leukemic cells is first choice, newer preclinical models including "organoids" and combinations of pharmacologic and genetic approaches may better align models with human AML. We advocate earlier inclusion of combinations +/- chemotherapy and of newly diagnosed patients into clinical trials. When a drug plausibly targets a pathway uniquely related to a specific genetic aberration, eligibility should begin with this subset, including patients with other malignancies, with subsequent extension to other patients. In other cases, a more open-minded approach to initial eligibility would facilitate quicker identification of responsive subsets. Complete remission without minimal residual disease seems a particularly useful short-term end point. Genotypic and phenotypic studies should be prespecified and performed routinely to distinguish responders from nonresponders.