Chemoradiotherapy and transplantation of bone marrow from matched sibling donors have been useful for the treatment of acute lymphoblastic leukemia in patients with a poor prognosis but are not available to some two thirds of patients who do not have a matched allogeneic donor. We undertook this study to compare autologous and allogeneic marrow transplantation in the treatment of such cases. We treated 91 patients with high-dose chemoradiotherapy and followed them for 1.4 to 5 years. Forty-six patients with an HLA-matched donor received allogeneic marrow, and 45 patients without a matched donor received their own marrow taken during remission and purged of leukemic cells with use of monoclonal antibodies. Bone marrow engraftment occurred earlier in patients who received autologous marrow. Recipients of autologous marrow had shorter hospital courses, with significantly fewer peritransplantation deaths than recipients of allogeneic marrow. Post-transplantation relapse of leukemia was the most frequent cause of treatment failure; relapses occurred in an estimated 37 percent of patients with allogeneic grafts in whom graft-versus-host disease developed, 75 percent of patients with allogeneic grafts in whom graft-versus-host disease did not develop, and 79 percent of patients who received autologous grafts. The interval before relapse was significantly shorter in the autologous-marrow group than in the allogeneic-marrow group. Recipients of autologous and allogeneic marrow whose first pretransplantation remissions were short (less than 18 months) had eventual outcomes similar to those whose first remissions were longer than 18 months. Patients with a first remission lasting less than 18 months had an outcome better than that expected with chemotherapy alone. The fractions of "cured" patients were estimated to be 20 percent in the autologous-marrow group and 27 percent in the allogeneic-marrow group--not a significant difference, but because of the limited statistical power of the study, the question of long-term disease-free survival must still be considered open.