To assess the accuracy of positron emission tomography (PET) for evaluation of coronary artery disease (CAD), cardiac PET perfusion images were obtained at rest and with dipyridamole-handgrip stress in 193 patients undergoing coronary arteriography. PET images were reviewed by two independent readers blinded to clinical data. Subjective defect severity scores were assigned to each myocardial region on a 0 (normal) to 5 (severe) scale. Results were compared with arteriographic stenosis severity expressed as stenosis flow reserve (SFR), with continuous values ranging from 0 (total occlusion) to 5 (normal), calculated from quantitative arteriographic dimensions using automated detection of the vessel borders. There were 115 patients with significant CAD (SFR less than 3), 37 patients with mild CAD (3 less than or equal to SFR less than 4), and 41 patients with essentially normal coronaries (SFR greater than or equal to 4). With increasingly severe impairment of stenosis flow reserve, subjective PET defect severity increased. Despite wide scatter, a PET score of 2 or more was highly predictive of significant flow reserve impairment (SFR less than 3). For each patient, the score of the most severe PET defect correlated with the SFR of that patient's most severe stenosis (rs = 0.77 +/- 0.06). For each of 243 stenoses, PET defect score correlated with the SFR of the corresponding artery (rs = 0.63 +/- 0.08). PET defect location closely matched the region supplied by the diseased artery, and readers agreed whether the most severe PET defect was less than or more than 2 for 89% of patients.