We treated 103 patients with histologically confirmed anaplastic supratentorial astrocytic neoplasms with either diaziquone (AZQ) and carmustine (BCNU) or AZQ and procarbazine. There were 74 patients with glioblastoma multiforme (GBM) and 29 patients with anaplastic astrocytoma (AA). AZQ plus BCNU produced partial (PR) or unequivocal responses in seven of 32 (21.9%) patients with GBMs and three of ten (30%) patients with AAs. Two patients with GBMs (6.3%) and five patients with AAs (50%) showed stable disease (SD). AZQ plus procarbazine produced PRs or unequivocal responses in five of 42 (11.9%) patients with GBMs and nine of 19 (47.4%) patients with AAs. Eight patients with GBMs (19%) and one patient with an AA (5.2%) showed SD. In addition to histologic diagnosis, only the Karnofsky performance-status (KPS) rating independently influenced response and survival. Differences in response rates between the two regimens were not significant, although estimated median survival after adjusting for performance status was slightly better with AZQ plus BCNU than with AZQ plus procarbazine (P = .031). Neither age nor prior chemotherapy were significant independent risk factors. Toxicity was mild and primarily hematologic. We conclude that these AZQ-based regimens have activity in patients with recurrent anaplastic gliomas, but that they are not clearly superior to other agents in current use. The histologic diagnosis of GBM is associated with a significantly worse prognosis than AA, and we believe that this important distinction must be recognized in phase II as well as phase III trials.