The use of immunodeficient mice, particularly of the nonobese diabetic/severe combined immunodeficient (NOD/SCID) strain, has allowed detection of very primitive malignant progenitors from patients with acute myelogenous leukemia (AML). To define the sensitivity and reproducibility with which the engraftment of different AML cells can be detected, 61 different samples from patients with newly diagnosed AML representing a variety of cytogenetic and French-American-British (FAB) subtypes were injected into NOD/SCID mice. Eight weeks after intravenous injection of 10⁷ AML cells, the average percent of human cells in mouse bone marrow was 13.3%, with 70% of samples showing easily detectable engraftment of CD45⁺ cells. AML samples with cytogenetic changes associated with a poor clinical prognosis tended to engraft to higher levels than those with changes associated with a good prognosis. Cells with FAB subtypes M3 and, to a lesser extent, M2, engrafted more poorly (P = .002 and .06, respectively) than those from other subtypes. Intraperitoneal injection of human interleukin-3 and Steel factor thrice weekly for 4 weeks did not enhance the levels of AML cell engraftment. However, AML samples that showed cytokine-independent colony growth in methylcellulose assay or expressed growth-factor mRNA in malignant blasts achieved significantly higher levels of engraftment than those which were cytokine dependent in culture or failed to express cytokine message (P < .03 and P < .02, respectively). In 6 patient samples, the frequency of NOD/SCID leukemia-initiating cells (NOD/SL-IC) varied from 0.7 to 45 per 10⁷ cells, which was 200- to 800-fold lower than the frequency of AML long-term culture-initiating cells (AML LTC-IC) in the same samples. Each NOD/SL-IC will produce more than 10⁶leukemic blasts as well as many AML-CFC and AML LTC-IC as detected 8 weeks postinjection into mice. Serial transplant experiments showed the ability of NOD/SL-IC to maintain their own numbers over at least 3 to 4 weeks in vivo. The ability of these progenitors to self-renew combined with their potential to differentiate to produce large numbers of more mature progenitors and leukemic blasts suggests that the NOD/SL-IC assay identifies leukemic ‘stem cells’ that may maintain the malignant clone in human patients. The further use of this assay should facilitate studies of AML stem cell biology and the evolution of novel therapeutic strategies.