This paper reports the clinical significance of a computer-analyzing system to detect and characterize interstitial lung diseases in chest radiographs. One hundred and sixty-four ROIs were selected in the right lungs of 41 patients with normal and those of 41 with diffuse interstitial involvement proved by X-ray CT. Selected ROIs were processed by 4-directional Laplacian-Gaussian filtering, binarization, and determination of linear shadows. For quantitative analysis of interstitial shadows, radiographic index, normalized percent-area of shadows in a ROI, was determined and evaluated in the images. Then, the radiographic indices were compared with CT-documented characteristics of interstitial lung shadows. The results were as follows: 1) Abnormal and normal lungs were well differentiated each other by all kinds of the radiographic indices obtained from the images filtered by 4-directional Laplacian-Gaussian filters and from those processed by determination of linear shadows. 2) ROIs with honeycombing shadows and with other interstitial shadows (interstitial changes other than honeycombing and nodulation) shown in CT were differentiated each other by the radiographic indices obtained from the summation image and the vertical directional image processed by determination of linear shadows (p less than .01). However, ROIs with multiple nodular shadows and with other interstitial shadows were not classified by these radiographic indices. These results indicate that this system may be useful for detection and characterization of interstitial diseases in chest radiographs.