Duplex ultrasonography is frequently used for noninvasive screening for extracranial occlusive disease. In a multicenter, prospective, randomized trial for carotid endarterectomy in symptomatic patients, the duplex ultrasound examination was compared with conventional angiographic findings for internal carotid artery (ICA) stenosis in 178 patients. Angiograms were interpreted by radiologists both at local medical centers and at a central site; the angiographic determination of ICA stenosis was calculated as percent diameter at the point of maximal narrowing compared with the normal distal ICA. Comparisons were made for 328 arteries, including both the symptomatic (> 50% stenosis determined by angiography) and the asymptomatic (variable degrees of stenosis) sides. Duplex ultrasonography sensitivity varied from 0.24 for 30 to 49% stenosis to 0.71 for 50 to 79% stenosis and 0.91 for ICA occlusion. Using a 50% stenosis cutpoint, duplex ultrasonography sensitivity was 0.90 with a specificity of 0.76. Duplex scan readings underestimated the degree of stenosis in the 30 to 49% stenosis group in 48% of the cases. There was no apparent relationship between the accuracy of stenosis determinations and that of external carotid artery stenosis, carotid plaque morphology, or ulceration determinations by ultrasound. On the basis of the benefit provided by carotid endarterectomy in symptomatic patients with high-grade lesions, duplex ultrasound accuracy is essential if noninvasive testing is used to make clinical decisions. In situations in which duplex findings may not be reliable, such as in the mild-to-moderate stenosis and occlusion categories, carotid angiography may be indicated.