Chemotherapy for brain tumors remains unsatisfactory. Despite increasing participation in clinical trials, there is a clear need for pre-clinical models. Heterotransplantation of surgical specimens directly into the anterior chamber of the nude mouse eye has been demonstrated to produce evaluable xenografts. Drug access in this model is considered to mimic the blood-brain barrier. Five clinical specimens in 3 children with primitive neuroectodermal tumor/medulloblastoma were the sources of 293 intraocular xenografts (5 cohorts by source). Each tumor-bearing mouse received 1 of 5 drugs or normal saline, by intraperitoneal injection, weekly for 5 weeks. Response was monitored for up to 22 weeks, using a staging system which estimates the proportion of the anterior chamber filled by tumor. Results were analysed both as response rates (shrinkage in excess of 50%) at the conclusion of the treatment course and as time to tumor progression by the life table method. Comparison of response rates within cohorts by source of xenografts (exact chi-square test for overall and 2-sided Fisher's exact test for paired comparisons) indicated cyclophosphamide to be the most effective single agent. In logrank analyses cyclophosphamide achieved significantly longer delays to progression than all other drugs in one cohort and longer delays than all but diaziquone in 2 other cohorts. The intraocular xenograft model is a clinically relevant system for the study of therapeutic agents in brain tumors. The effectiveness of intensive dosage cyclophosphamide in a model dependent on access across the blood-aqueous barrier is important and consistent with recent clinical data.