This study was designed to elucidate the operator- and patient-dependent variables inherent in clinical application of quantitative coronary arteriography. Digital arteriograms from 25 consecutive patients undergoing diagnostic catheterization were analyzed by four experienced angiographers utilizing an automated coronary edge detection system to measure percent area stenosis. The identification of potentially significant lesions for quantitation constituted a major source of variability, with unanimous agreement on the presence of a greater than or equal to 50% stenosis occurring at 38 (29%) of the 130 reported sites. Selection of an optimal frame for quantitative analysis resulted in disagreement for every lesion reported. Frame selection by the operator, as opposed to measurement of preselected frames, increased the interobserver variability from 5% to 7% for automated geometric analysis (p less than 0.01), and from 8% to 10.5% for automated densitometric analysis (p less than 0.01). Fully automatic arterial border detection was possible for only 20 (52.5%) of the 38 unanimously identified stenoses. The 18 failures involved one or more of the following factors: 1) stenosis at a bifurcation (13 [72%]); 2) diffuse, severe disease (8 [44%]); 3) excessive vessel tortuosity or overlap or both (4 [22%]); and 4) poor image quality (5 [28%]). In contrast, the same automated border detection algorithm successfully traced all 15 preselected frames of discrete stenoses referred for coronary angioplasty. Automated quantitative coronary arteriography performs well when carefully selected, discrete stenoses are presented to the computer for analysis. However, quantitative analysis of routine clinical coronary arteriograms is limited by operator-dependent variability in stenosis identification and frame selection, as well as by complex coronary anatomy and suboptimal image quality. These limitations make automated quantitative coronary arteriography impractical for routine clinical use.