OBJECTIVE To determine the accuracy of magnetic resonance angiography in assessing patients with cerebrovascular disease, we performed a study comparing the results of conventional cerebral angiography, duplex scanning, and magnetic resonance angiography. METHODS From 42 patients, a total of 25 carotid arteries were evaluated by all three techniques. The studies were independently read and sorted into five categories according to the degree of stenosis: 0-15%, normal; 16-49%, mild; 50-79%, moderate; 80-99%, severe; and totally occluded. RESULTS Magnetic resonance angiography correlated exactly with conventional angiography in 39 arteries (52%); duplex scanning correlated with conventional angiography in 49 cases (65%). Compared with conventional angiography, both magnetic resonance angiography and duplex scanning tended to overread the degree of stenosis. The most critical errors associated with magnetic resonance angiography were three readings of total occlusion in vessels found to be patent on conventional angiograms. CONCLUSIONS Although magnetic resonance angiography offers great hope of providing high-quality imaging of the carotid artery with no risk and at less cost, data from this study suggest that misreading the degree of stenosis, or misinterpreting a stenosis for an occlusion, could lead to errors in clinical decisions. Guidelines for use of magnetic resonance angiography in a clinical setting are offered.