The purpose of this study was to compare transplantation outcomes in patients with hematologic malignancies who received marrow grafts from either phenotypically matched unrelated, one- antigen–mismatched unrelated, or highly human leukocyte antigen (HLA)–disparate family donors. Between 1993 and 2000, 139 patients underwent transplantation from unrelated donors (81 matched and 58 mismatched) and 48 patients received marrow grafts from family donors that were mismatched at 2, 3, or 4 of 8 HLA loci. All patients received a standardized conditioning regimen and a graft-versus-host disease (GVHD) prophylaxis schedule with the exception of recipients of haploidentical marrow grafts, who received antithymocyte globulin after bone marrow transplantation as additional immunosuppression. There was no statistically significant difference in the rate of engraftment, or the cumulative incidences of acute and chronic GVHD between any of the 3 groups. The 2-year cumulative incidence of relapse was lower in matched unrelated patients (25%, P = .01) and mismatched unrelated patients (26%,P = .014) than in haploidentical patients (42%). Transplant-related mortality was significantly higher in recipients of mismatched unrelated grafts (45%, P = .01) and haploidentical grafts (42%, P = .001) compared with recipients of matched unrelated marrow grafts (23%). This resulted in a significantly higher probability of overall survival for matched unrelated patients (58%) versus either mismatched unrelated (34%,P = .01) or haploidentical (21%, P = .002) patients. There was no statistically significant difference in survival between patients who received mismatched unrelated grafts versus those who received haploidentical grafts. This study supports a donor selection algorithm whereby patients who lack a closely matched family donor be offered a phenotypically matched unrelated donor if available. There is no apparent advantage to using a mismatched unrelated versus a highly HLA-disparate family donor.