To accurately diagnose stenotic lesions on coronary cineangiograms, an automatic detection method using computer image processing was developed. We evaluated its accuracy by comparing the results of computer-aided interpretation (CAI) with those obtained independently by 3 observers. Evaluation was performed on 129 segments from 27 arteries visualized on angiograms obtained in 18 patients. The detection rates of stenosis of the 3 observers by pure visual interpretation were 7.0%, 27.9%, and 17.1%, and using CAI 40.0%, 42.6%, and 47.3%. By computer recognition alone, a detection rate of 51.9% was achieved. The agreement by at least 2 observers (consensus) on the sites with lesions was 41.1% while the consensus of computer recognition regarding the sites with lesion was 40.3%. Therefore, our findings indicated that computer recognition of cineangiograms is likely to result in overdetection of lesions. However, all 3 observers detected stenotic lesions better with CAI than with pure visual interpretation. Accordingly, CAI may improve the reliability of cineangiographic diagnosis.