Predicting behavior based on histologic appearance has been problematic in ependymomas. Sixty-one patients with ependymoma (excluding subependymoma and myxopapillary ependymoma) were studied. The patients included 36 men and ranged in age from 1.5 to 74 years (median, 33 years). The most common clinical presentations included headache (n = 19), weakness (n = 18), nausea/vomiting (n = 12), and gait disturbance (n = 10). Location included spinal cord (n = 24), fourth ventricle (n = 21), lateral ventricle (n = 8), and third ventricle (n = 5). Initial surgery included a gross total resection of tumor in 22 patients and subtotal resection or biopsy in the remaining patients. Thirty-five patients were known to have been treated with adjuvant radiation therapy and 13 patients received adjuvant chemotherapy. At last known follow-up, 20 patients were alive with no evidence of tumor (median, 66.5 months), 17 patients were alive with residual tumor (median, 14 months), and 12 patients died of tumor (median, 27.5 months). Two additional patients are alive with tumor status not known, two cases are current, and two patients were lost to follow-up. The additional six patients died either shortly after surgery or of surgical complications. Sixteen of 18 patients had at least one tumor recurrence at median 28.5 months. Fifty-one tumors had a predominantly glial pattern and 10 had a mixed glial-epithelial pattern. Of histologic features examined, patients with tumor recurrence or who died of tumor more frequently had observable mitotic figures, vascular proliferation, necrosis, and foci of increased cellularity. Eight of 18 recurrent tumors were classified as high grade ependymomas (anaplastic/malignant). Of patients who died of tumor, 4 of 12 had histologically high grade tumors versus 5 of 39 of the remaining tumors. MIB-1 immunostaining (marker of cell proliferation) was performed on 50 tumors. MIB-1 labeling indices (% positive tumor cell nuclei) ranged from 0.1 to 34.0 (median, 1.1). A higher percentage of patients with recurrent tumor (6 of 13, 46%) or who died of tumor (3 of 10, 30%) had MIB-1 indices >/= 4.0 versus the remaining patients (8 of 33, 24%). The conclusions are as follows: (1) histologic appearance and MIB-1 indices were not reliably predictive of tumor behavior, probably due in part to tumor heterogeneity; (2) tumors with two or more of the following features: identifiable mitotic figures, hypercellularity, vascular proliferation, and necrosis were more likely to behave in an aggressive manner; and (3) elevated MIB-1 labeling indices (>/=4.0 in this study) were encountered in a higher percentage of fatal and recurrent tumors than in nonfatal or nonrecurrent tumors.