Purpose: To assess the response of patients with recurrent malignant gliomas to intra-arterial (IA) cisplatin. Methods: Eligibility criteria included patients with recurrent supratentorial malignant gliomas and measurable, unilateral contrast-enhancing tumor located within the territory of one or two major cerebral arteries. Patients received 75 mg/m2 IA cisplatin every four weeks. Depending on individual patients' tumor topography, cisplatin was infused either into the cervical internal carotid artery (ICA) (15 patients), or into one or two major cerebral arteries (26 patients), most often the M1 segment of the middle cerebral artery. Results: Of 40 patients evaluable for tumor response, four patients (10%) were responders and nine patients (22%) had disease stabilization. The median time to tumor progression among the 13 patients with tumor response or stable disease was 23.7 weeks. The response rate did not significantly differ between patients receiving ICA versus selective intracerebral infusion, although the latter group contained a higher proportion of glioblastoma. Tumor progression occurred solely as local failure in 33 patients (82%), with all enhancing tumor still located within the vascular territory infused with IA cisplatin. Ipsilateral vision loss occurred in two patients after ICA cisplatin but in none of the selective infusion patients. Seizures and/or transient or permanent neurologic deterioration occurred in four patients (27%) after ICA cisplatin and in 11 patients (44%) after selective intracerebral infusion. Conclusions: Although this was not a randomized comparison, selective intracerebral artery cisplatin infusion in this group of patients reduced the risk of eye toxicity, but did not produce a better tumor response rate, and carried a higher risk of neurotoxicity relative to ICA infusion.