To develop a purification strategy for isolating the most primitive hematopoietic stem cells present in normal human marrow we have combined cell separation techniques with an assay for cells that initiate sustained hematopoiesis in vitro in the presence of irradiated human marrow adherent cells. These "feeders" were established by subculturing 2- to 6-week-old primary long-term marrow culture adherent layers at a density of 3 x 10(4) irradiated cells per square centimeter. Test "long-term culture (LTC)-initiating cells" were plated on top of the feeders and the cocultures then maintained as standard long-term marrow cultures with half-media changes and removal of half of the nonadherent cells each week. The total number of myeloid, erythroid, and multilineage clonogenic progenitors present after 5 weeks was used to provide a quantitative assessment of the number of LTC-initiating cells originally added. Using this assay, the density, light scatter, and two cell surface antigen properties of LTC-initiating cells have been defined and compared with cells capable of directly forming colonies in methylcellulose. While the majority of the clonogenic cells were found in the high forward light scatter (FLS) "blast" window, LTC-initiating cells had significantly lower FLS properties and in this respect were more similar to lymphocytes. LTC-initiating cells also expressed less HLA-DR antigen than clonogenic cells. The majority of LTC-initiating cells were found in the top 2% of the CD34 (My10) fluorescence profile, whereas clonogenic cells were found throughout the top 5% of the CD34 fluorescence profile. By combining low FLS, low orthogonal light scatter (OLS), low HLA-DR expression, and high CD34 expression, a population could be obtained that was enriched for LTC-initiating cells approximately 800-fold over unseparated marrow. This population contains only 0.06% of the marrow cells and 2% of the total clonogenic cells, but retains 50% to 60% of the LTC-initiating cells present in the original marrow. The ability to purify these two populations independently shows that the LTC and clonogenic assays identify distinct, although not necessarily nonoverlapping cell types in human marrow. Since clonogenic cells are derived from LTC-initiating cells, the LTC assay clearly detects a more primitive population. The availability of a simple approach that allows the purification of such cells by three orders of magnitude in high yield should be useful for the investigation of early events in hematopoiesis as well as for the definitive isolation of human hematopoietic stem cells with long-term in vivo repopulating potential.