Duplex scanning has been advocated as an acceptable alternative to angiography in the preoperative evaluation of carotid artery stenosis. To evaluate the accuracy of carotid Doppler in differentiating severe carotid stenosis from occlusion, we compared the results of angiography with duplex scanning in 124 carotid arteries (62 patients) and with continuous-wave Doppler in 662 carotid arteries (331 patients). The specificity was 95-99%, sensitivity was 86-96%, and accuracy was 95-98%. Duplex scanning wrongly identified occlusion in four arteries and failed to detect occlusion in one artery. In making decisions prior to carotid endarterectomy, even infrequent errors are unacceptable. We recommend angiography of all surgical candidates with apparent severe stenosis when the internal carotid artery cannot be clearly identified on duplex, or to distinguish apparent occlusion from undetectably low blood flow.