Currently, automated systems for quantitative analysis by intracoronary ultrasound (ICUS) are restricted to the detection of the lumen. The aim of this study was to determine the accuracy and reproducibility of a new semiautomated contour detection method, providing off-line identification of the intimal leading edge and external contour of the vessel in three-dimensional ICUS. The system allows cross-sectional and volumetric quantification of lumen and of plaque. It applies a minimum-cost algorithm and the concept that edge points derived from previously detected longitudinal contours guide and facilitate the contour detection in the cross-sectional images. A tubular phantom with segments of various luminal dimensions was examined in vitro during five catheter pull-backs (1 mm/sec), and subsequently 20 diseased human coronary arteries were studied in vivo with 2.9F 30 MHz mechanical ultrasound catheters (200 images per 20 mm segment). The ICUS measurements of phantom lumen area and volume revealed a high correlation with the true phantom areas and volumes (r = 0.99); relative mean differences were -0.65% to 3.86% for the areas and 0.25% to 1.72% for the volumes of the various segments. Intraob-server and interobserver comparisons showed high correlations (r = 0.95 to 0.98 for area and r = 0.99 for volume) and small mean relative differences (-0.87% to 1.08%), with SD of lumen, plaque, and total vessel measurements not exceeding 7.28%, 10.81%, and 4.44% (area) and 2.66%, 2.81%, and 0.67% (volume), respectively. Thus the proposed analysis system provided accurate measurements of phantom dimensions and can be used to perform highly reproducible area and volume measurements in three-dimensional ICUS in vivo.