Digital analysis of 80 plain chest radiographs was done to differentiate benign from malignant pulmonary nodules. Each of 80 radiographs had a solitary pulmonary nodule, of which 40 were benign and the other 40 were malignant. Radiographic features of nodules were visually evaluated. Frequencies of the radiographic features described below were significantly different between benign and malignant nodules: irregular shape, vague density, well-defined contour, spicular formation, and vascular convergene. To digitally evaluate the density characteristics, two kinds of concentric circular windows were used, to allow analysis of intra-nodular, marginal and surrounding regions of the nodules. Seventeen digital parameters were obtained by computerized analysis with the windows. There were significant differences in six digital parameters; three of density gradient (one of intra-nodular and two of marginal regions), one DCF-N (directional contrast filter for nodule) output of intra-nodular region, and two of density entropy values of intra-nodular and surrounding regions. Density gradient parameters of marginal regions are believed to represent the features of nodular shape and spicula formation. With these six parameters, 70% of benign nodules and 85% of malignant nodules were correctly discriminated. These results suggest that computer-aided diagnosis may be used to differentiate benign from malignant nodules.