Factors contributing to the development of central nervous system (CNS) leukemia, and the impact of leukemic involvement of this site on subsequent remission length, were determined in 184 children with acute nonlymphoblastic leukemia who had been treated in two successive clinical trials. Preventive CNS therapy in both studies consisted of intrathecal methotrexate (12 mg/m2) given monthly during the first six months of therapy and then every three months until all treatment was stopped. Children with CNS leukemia at diagnosis or relapse were given intrathecal chemotherapy weekly for four weeks and then monthly throughout the remainder of the treatment course. Those continuing in complete remission received 2,400 rad cranial irradiation plus five doses of intrathecal methotrexate before cessation of therapy. The 38 children (20.7%) with CNS leukemia at diagnosis were more likely to have an initial leukocyte count greater than or equal to 25 X 10(9)/L (P = .01) and age less than 2 years (P = .03). The presence of CNS leukemia at diagnosis did not adversely affect the remission induction rate (P = .13) or the length of complete remissions (P = .73). CNS relapse ended initial remissions in 11 patients only and did not preclude subsequent long-term survival, as four of these children are off therapy and in second complete remission for 33+ to 78+ months. Three features at diagnosis were predictive of CNS relapse: monocytic or myelomonocytic leukemia (P = .002); age less than 2 years (P = .0001); and leukocyte count greater than or equal to 25 X 10(9)/L (P = .012). By stepwise Cox regression analysis, each factor was found to have independent predictive value. Despite the apparent effectiveness of intrathecal methotrexate as preventive CNS treatment, our findings indicate that more effective prophylaxis is needed for patients with features predisposing to CNS relapse.