Eight hundred seventy arteriograms from the Coronary Artery Surgery Study (CASS) were independently read by readers at two different clinics to evaluate the reproducibility of the interpretation of coronary arteriograms. Among proximal segments, the interpretation of lesions of the left main coronary artery were the least reproducible, P less than .02. When one angiographer reads a stenosis of 50% or more in the left main coronary artery, it is estimated that a second reader will report no lesion 18.6% of the time. In 94.7% of the films, the number of significantly (greater than or equal to 70% stenosis) diseased vessels was the same for both readers (72.1%) or differed by one vessel (22.6%). The reproducibility of interpretation of films of good or acceptable quality or completeness was better than the reproducibility of readings of arteriograms judged to be of poor quality or incomplete studies. The mean absolute difference between readings of the percent stenosis decreased over the time of the patient enrollment, 1975 to 1978. This may have resulted from major collaborative efforts made during the course of the study to improve the quality of angiography and to standardize the reading of the cine films.