An ROC study is described which compares the performance of three types of images--conventional screen-film, single-energy digital and dual energy bone cancelled (soft tissue) digital--in detecting subtle interstitial pulmonary disease. Marginally detectable nodular and reticulonodular patterns (12 different patterns of each) were superimposed over the lungs of a frozen human chest phantom to simulate the clinical situation. The digital images were formatted on film at full size (ie, 35 cm X 43 cm). A total of 156 images (52 of each type, of which 28 were normal and 24 had simulated pathology) were used in the study and read by five experienced chest radiologists. Using a paired t-test, the areas under the individual ROC curves were compared for three combinations of images--single-energy digital and conventional, soft tissue digital and conventional, and soft tissue and single-energy digital. No statistically significant difference was observed between the conventional and single-energy digital images. The readers performed better with both conventional and single-energy digital images than with the soft tissue digital images at statistically significant levels (P = 0.05 for conventional vs. soft tissue digital and P = 0.02 for single-energy digital vs. soft tissue digital). The results suggest that there is no advantage in employing dual-energy soft tissue images to assist in diagnosing interstitial disease in the clinical setting. They also suggest that spatial resolution requirements are less demanding in digital chest systems that obtain scatter-free images than in digital systems utilizing conventional scatter control techniques.