A computer program that recognizes potential pulmonary nodules in PA chest radiographs has been developed. This program produces a display of candidate nodules that require interpretation by a radiologist. Some false positives are rejected by a program, the Nodule Expert. Detection performance with and without Nodule Expert has been evaluated. Using the untrained program (no Nodule Expert), and after inspecting 45 candidate nodules, a radiologist may be confident that a nodule was inspected, if one was located by the program. When pattern recognition techniques are incorporated, the number of false positives presented for inspection is reduced. The radiologist must inspect, at most, 10 candidate nodules to be confident of having inspected a nodule, if one was located by the program. Concomitant with this decrease in the candidate nodule false-positive rate is a decrease in sensitivity (film true-positive rate) from 92 to 86%. This program was trained on candidate nodules from 37 radiographs and also tested on these 37. Some of the features used by the pattern classifier to classify candidate nodules are comparable to those used by human observers.