Digital video subtraction angiography (DVSA) and duplex scanning were compared to conventional angiography with regard to their accuracy in quantifying internal carotid artery disease. A classification of 108 carotid arteries available for comparison was made into five groups: normal (14), stenosis to 20% (15), stenosis from 20% to 49% (19), stenosis from 50% to 99% (44), and occlusion (16). The overall agreement corrected for chance (K) for DVSA was 0.738 +/- (SE) 0.049, which was better than for duplex scanning (K = 0.610 +/- 0.055). For hemodynamically significant disease (greater than or equal to 50% diameter reduction), both techniques were highly accurate: DVSA had a sensitivity of 95% (57/60) and a specificity of 92% (44/48); the sensitivity of duplex scanning was also 95% (57/60) and the specificity 85% (41/48). The accuracy in differentiating between high-grade stenosis and occlusion was 98% (59/60) for duplex scanning and 100% (60/60) for DVSA. The most prominent limitation of duplex scanning was its overestimation of disease in normal and minimally diseased arteries. Its specificity for ruling out any degree of disease was only 21% (3/14). DVSA had a poor predictive value of a normal test, 55% (12/22), and was more than one category wrong seven times, while this occurred only three times for duplex scanning. At the present time carotid endarterectomy without conventional angiography is only advocated when there is complete agreement between DVSA and duplex scanning.