Eighty-seven children with central nervous system (CNS) leukemia were randomized to receive either induction intrathecal chemotherapy (ITC) and cranial irradiation (CRT) plus maintenance ITC, or induction ITC and craniospinal irradiation (CSpRT) with no maintenance ITC. ITC consisted of six weekly injections of methotrexate, hydrocortisone, and arabinosylcytosine. Also, intensification of systemic induction and maintenance chemotherapy was given. CRT + ITC was given as CRT, 2400 rad in 12 fractions followed by ITC maintenance bimonthly for 2 years. Craniospinal irradiation consisted of CRT + 1400 rad in ten fractions to the spine. Randomization was stratified according to whether CNS leukemia occurred at initial diagnosis of acute lymphocytic leukemia (ALL) (Stratum I, 15 patients), during first bone marrow (BM) remission (Stratum II, 49 patients), simultaneous with first BM relapse (Stratum III, 12 patients), or during second BM remission (Stratum IV, 11 patients). The median follow-up for patients who remain at risk is 15 + months. Eight children (seven on CRT + ITC, one on CSpRT) developed presumed therapy related encephalopathy. In Stratum II, 16 of 29 (55%) patients receiving CRT + ITC experienced adverse events: 3 deaths during continuous complete remission (CCR) and 13 relapses (2 CNS, 1 CNS + BM, 1 BM + testes, and 2 testes) as compared with only 5 relapses in 20 (25%) patients on CSpRT (1 CNS, 1 CNS + BM, 1 BM, and 2 testes). The children on both regimens were comparable for sex, race, age at initial ALL diagnosis, time from ALL diagnosis to first episode of CNS leukemia, systemic therapy both before and after CNS relapse, and number of blasts in the spinal fluid at diagnosis of CNS leukemia. The conclusion is that children with isolated CNS leukemia can achieve prolonged survival with aggressive therapy, and that CSpRT is possibly less toxic and more likely than is CRT + ITC to prevent subsequent BM and testicular relapse (P less than 0.02), but not subsequent CNS relapse (P = 0.7). A possible systemic therapy effect of spinal irradiation is postulated to explain the superiority of CSpRT.