In a retrospective study of 48 patients who underwent elective surgery for cerebral arteriovenous malformations, a statistical analysis of demographic, clinical, and neuroradiological data was undertaken in order to discover the best predictors of operative morbidity. In addition, the predictive value of different clinical grading systems as applied to this series was compared. All patients had a computed tomographic scan and a positive angiogram before surgery. Complete resection was proven angiographically. The univariate Mann-Whitney-Wilcoxon rank sum test, the Fisher exact test, Spearman's rank correlation coefficient analysis, and multivariate logistic regression were used as statistical methods. Duration of surgery, the development, of either new deficits or an increase in the preoperative neurological signs immediately after surgery, and rehabilitation (as measured by the Karnofsky index) were taken as target variables for the difficulty of operation and for postoperative morbidity, respectively. The largest diameter of the nidus of the arteriovenous malformation, eloquence of the adjacent brain, and deep venous drainage showed the most consistent correlation with these target variables. Intracerebral hematoma and other single factors, such as the age of the patient or localization of the arteriovenous malformation did not affect the outcome. The clinical grading scale of Spetzler and Martin provided better prediction of surgical risks than other proposed systems.