Variability in coronary arteriogram readings was studied by having cine films from ten patients read by eleven readers. Three of the eleven subsequently met as an expert panel to provide a joint evaluation which could serve as a standard. Considerable variability was found between individual readers and between readers and the panel. The average standard deviation for estimation of any segmental stenosis by any single reader was 18%. Disagreement about the number of major vessels with a 70% stenosis occurred 31% of the time. Discrepancies were most likely to occur in analyzing distal arterial segments, in reading nonopacified segments, and during analysis of films showing more severe disease or having poorer technical quality. Recent experience in reading arteriograms seemed to be the most important characteristic in determining the accuracy of a reader. A protocol for the use of three readers is suggested.