Computed tomography (CT) and mediastinoscopy were compared in 151 patients with bronchogenic carcinoma. In all patients in whom findings at mediastinoscopy were negative, all accessible nodes were either removed or sampled at thoracotomy. Several size criteria for identifying nodes as enlarged on CT scans were compared. The long axis greater than or equal to 15 mm and short axis greater than 10 mm had very low sensitivity (61%), and the long axis greater than 5 mm had a low specificity (23%). CT (long axis greater than 10 mm) allowed sensitivity equal to that of mediastinoscopy (79%) in the detection of mediastinal metastases, but the specificity with CT was lower (65% vs. 100%). In seven of 44 patients with nodes greater than 10 mm on CT scans and with positive findings at mediastinoscopy, tumor was present not in the enlarged nodes but rather in normal-sized nodes in a different nodal station. The sensitivity of CT for actual nodal stations involved with tumor was only 66%. Eighty-three percent of patients with false-negative findings at mediastinoscopy but only 33% of patients with false-negative findings at CT had surgically resectable stage IIIa disease.