The decision-making process whereby treatment is offered to a patient with an arteriovenous malformation (AVM) must be supported by an understanding of the risks related to the natural history of the AVM and the risks related to the treatment of that particular AVM. The ability to estimate the treatment risk for an individual patient is hampered by the marked variability in the complexity of AVMs. In 1986, an AVM grading system was proposed to predict surgical morbidity and mortality. This system is based on the AVM size, the neurological eloquence of adjacent brain, and the pattern of venous drainage. Grade I malformations are small, superficial, and located in noneloquent cortex; Grade V lesions are large, deep, and situated in neurologically critical areas; and Grade VI lesions are considered inoperable AVMs. A retrospective application of this grading scheme demonstrated its correlation with the incidence of postoperative neurological complications. A prospective application of the AVM grading system has been performed in 120 consecutive patients who had a complete microsurgical excision of their AVM, with or without AVM embolization. The AVM grading system accurately correlated with both new-temporary (P < 0.0001) and new-permanent (P = 0.008) neurological deficits. The permanent major neurological morbidity rates for Grades I through III were 0%, increasing to 21.9% in patients with Grade IV and 16.7% in patients with Grade V AVMs (P < 0.0001). One patient with a Grade III AVM died from an esophageal hemorrhage 15 months after her AVM was treated.(ABSTRACT TRUNCATED AT 250 WORDS)