Children with acute lymphoblastic leukemia (ALL) who have completed 2.5 to 3 years of initial chemotherapy have an off-therapy relapse rate of approximately 20%. In an attempt to improve the survival of children with a late bone marrow (BM) relapse (ie, occurring greater than 6 months after cessation of primary therapy), the Pediatric Oncology Group designed a randomized study to compare the efficacy of doxorubicin/prednisone and cytarabine/teniposide in a multidrug retreatment chemotherapy program. Treatment consisted of remission reinduction with vincristine, prednisone, and doxorubicin, central nervous system prophylaxis with triple intrathecal chemotherapy, and continuation therapy (for 132 weeks) with alternating cycles of oral 6-mercaptopurine/methotrexate and intravenous vincristine/cyclophosphamide. Patients received intermittent courses of either prednisone/doxorubicin (regimen 1) or teniposide/cytarabine (regimen 2) during continuation therapy and a late intensification phase with either vincristine, prednisone, and doxorubicin (regimen 1) or teniposide and cytarabine (regimen 2). One hundred two of 105 evaluable patients (97%) achieved a second complete remission. Twenty-eight of 50 patients on regimen 1 have failed compared with 28 or 52 patients on regimen 2 (log-rank analysis, P = .68), indicating that this trial was inconclusive as to which treatment regimen was superior. The overall 4-year event-free survival for children with a late BM relapse was 37% +/- 6%. Age less than 10 years at initial diagnosis (P < or = .001), white blood cell count less than 5,000/microL at relapse (P = .036) and duration of first remission greater than 54 months (P = .039) were independently associated with a more favorable outcome. While the randomized trial was inconclusive, prolonged second complete remissions were secured in more than one-third of children with a late BM relapse of ALL. The prognostic factors identified may help select children with a late BM relapse who can be successfully retreated with chemotherapy alone.