provide views at 900 angles to one another. No hemiaxial views were taken. A stratified random sampling scheme was used to ensure selection of a representative spectrum of coronary disease groups, as categorized by the number of major vessels with at least 70% stenosis and by normal or abnormal left ventricular function. Only films estimated to be of adequate technical quality were eligible for this study. Eleven physicians, representing the Seattle Heart Watch cooperating facilities, individually reviewed these ten films in a pre-assigned random order and completed the standard Heart Watch arteriography forms. Three of the eleven subsequently met to resolve differences of opinion and provide a joint reading which could serve as a standard from an "expert panel."Raw observations collected were: the estimated amount of stenosis (percent narrowing in the diameters) of each of ten vessel segments, the presence or absence of collaterals, the degree of left ventricular contraction according to a qualitative five-division grading, as well as a film quality rating, and an overall assessment of the presence of coronary artery disease.The statistical approach taken for all entirely numerical variables was to use the standard deviation of the responses as a measure of the amount of variability and thus of the amount of agreement. When considering the amount of agreement on variables that were at least partially nominal in nature, standard deviations were not always applicable. For example, each reader was asked to estimate the amount of narrowing (to the nearest 10%) in the diameters of each of ten vessel segments from each patient. However, when the vessel segment was not sufficiently visible to allow an assess-324 CIRCULATION by guest on