We compared the abilities of positron emission tomography and computed tomography to detect N2 or N3 lymph node metastases (N2 or N3) in patients with lung cancer. Positron emission tomography detects increased rates of glucose uptake, characteristic of malignant cells. Patients with peripheral tumors smaller than 2 cm and a normal mediastinum were ineligible. All patients underwent computed tomography, positron emission tomography, and surgical staging. The American Thoracic Society lymph node map was used. Computed and positron emission tomographic scans were read by separate radiologists blinded to surgical staging results. Lymph nodes were "positive" by computed tomography if larger than 1.0 cm in short-axis diameter. Standardized uptake values were recorded from areas on positron emission tomography corresponding to those from which biopsy specimens were taken; if greater than 4.2, they were called "positive." Seventy-five lymph node stations (2.8 per patient) were analyzed in 27 patients. Computed tomography incorrectly staged the mediastinum as positive for metastases in three patients and as negative for metastases in three patients. Sensitivity and specificity of computed tomographic scans were 67% and 83%, respectively. Positron emission tomography correctly staged the mediastinum in all 27 patients. When analyzed by individual node station, there were four false positive and four false negative results by computed tomography (sensitivity = 60%, specificity = 93%, positive predictive value = 60%). Positron emission tomography mislabeled one node station as positive (100% sensitive, 98% specific, positive predictive value 91%). The differences were significant when the data were analyzed both for individual lymph node stations (p = 0.039) and for patients (p = 0.031) (McNemar test). Positron emission tomography and computed tomography are more accurate than computed tomography alone in detecting mediastinal lymph node metastases from non-small-cell lung cancer.