Computer-assisted contour detection and videodensitometric cross sectional area assessment of coronary artery obstructions on the CAAS II system were validated in vitro and in vivo by angiographic cinefilm recording and automated measurement of stenosis phantoms (luminal diameter 0.5, 0.7, 1.0, 1.4, 1.9 mm) which were first inserted in a plexiglass model and then serially implanted in swine coronary arteries. "Obstruction diameter" (OD) and "obstruction area" (OA) values obtained from 10 in vitro and 19 in vivo images at the site of the artificial stenoses were compared with the true phantom dimensions. The in vitro assessment of OD yielded an accuracy of 0.00 +/- 0.11 mm (correlation coefficient: r = 0.98, y = 0.18 + 0.82x, standard error of estimate: SEE = 0.08), whereas the in vivo measurement of OD gave an accuracy of -0.01 +/- 0.18 mm (r = 0.94, y = 0.22 + 0.82x, SEE = 0.15). The assessment of OA gave an accuracy of -0.08 +/- 0.21 mm2 in vitro (r = 0.97, y = 0.08 + 0.99x, SEE = 0.22) and -0.22 +/- 0.32 mm2 in vivo (r = 0.95, y = 0.21 + 1.01x, SEE = 0.33). The mean reproducibility was +/- 0.09 mm for geometric measurements and +/- 0.21 mm2 for videodensitometric assessments, respectively. Thus, due to inherent limitations of the imaging chain, the reliability of geometric coronary measurements is still far superior to videodensitometric assessments of vessel cross sectional areas.