We conducted a phase I-II clinical trial to test the hypothesis that removal of CD4 cells from an HLA-mismatched unrelated marrow graft would substantially reduce the risk of grades III-IV graft-versus-host disease (GVHD) and that retention of a specified number of CD8 cells in the graft would be sufficient to prevent rejection. Patients were eligible for this study when an HLA-A, -B, or -DRB1–matched unrelated donor could not be identified. HLA matching of the donor and recipient was based on typing of HLA-A and -B antigens by serologic methods and by typing of HLA-DRB1 alleles by molecular methods, and donors were selected when disparity was limited to a single HLA-DRB1 allele or a single HLA-A or -B antigen. Twenty-seven patients with hematologic malignancy or aplastic anemia were prepared to receive a transplant with conventional regimens of cyclophosphamide and fractionated total body irradiation, and a standard regimen of methotrexate and cyclosporine was given for GVHD prophylaxis. CD4 cells were removed from the donor marrow, and the numbers of CD8 cells were adjusted systematically in graded steps for successive patients, depending on the occurrence of grades III-IV GVHD or graft failure in previously enrolled patients. Removal of CD4 cells did not cause graft rejection or appreciably decrease the risk of grades III-IV GVHD. Depletion of CD8 cells was associated with an increased risk of rejection with either HLA-DRB1 disparity or with HLA-A or -B disparity. With either type of disparity, the risk of grades III-IV GVHD is likely to be higher than 15% at any dose of CD8 cells associated with less than 5% risk of graft failure. The absence of graft failure associated with CD4 depletion supports the hypothesis that donor CD4 cells are not essential for preventing marrow graft rejection in humans. The correlation between graft failure and the number of CD8 cells in the donor marrow supports the hypothesis that donor CD8 cells help to prevent marrow graft rejection.