To test the reliability of conventional selective pulmonary arteriography in the diagnosis of pulmonary embolism, three angiographers reviewed the arteriograms of a series of 60 patients retrospectively, independently, and without benefit of additional data. Pulmonary arteriograms had been interpreted as positive for pulmonary embolism in 25 of these patients during their hospitalizations. Angiographers A, B, and C judged the arteriograms of 24, 29, and 25 patients, respectively, as positive for pulmonary embolism. Mean interobserver agreement was 86%. Interobserver agreement was not associated significantly with the quality of the arteriogram or with selective injection of a lobar vs a pulmonary artery, but was associated strongly with the magnitude of thromboembolism. All angiographers agreed that the arteriograms were positive in 18 cases of pulmonary embolism graded as massive, lobar, or segmental, but agreed in only two of 15 cases graded as subsegmental. We conclude that conventional selective pulmonary arteriography is reliable in the detection of embolus in segmental or larger pulmonary arteries. Observer disagreement becomes considerable for embolus limited to subsegmental pulmonary arteries, indicating that emboli of this size are at the resolution limit of the technique.