As part of our continuing evaluation of the clinical applicability of digital radiography, we compared the abilities of radiologists to detect pneumothoraces on conventional chest radiographs with their performances when using three formats of digitally obtained images. Twenty-three frontal-view chest radiographs with pneumothoraces and 22 other chest radiographs, either normal or showing miscellaneous abnormalities, were interpreted by five experienced radiologists in each of four formats: conventional film-screen chest radiographs, small-format (17.8 x 21.6 cm) computed radiographs, large-format (35.6 x 43.1 cm) computed radiographs, and digital images viewed on an interactive electronic workstation. The receiver-operating-characteristic curve areas for each observer for the four types of images were compared by a z test on a critical ratio, and the mean sensitivity and specificity values were compared by the sign rank test. The mean areas under the receiver-operating-characteristic curves ranged from 0.869 for the digital workstation to 0.915 for film-screen images. The differences observed among formats were not statistically significant. Mean specificities also were not significantly different, ranging from 0.90 for large-format computed radiographs to 0.96 for the digital workstation. Mean sensitivity ranged from 0.65 for the digital workstation to 0.82 for film-screen images. Radiologists interpreting digital workstation images were significantly less sensitive in detecting pneumothoraces than with film-screen and small-format computed images (p = .06). In this study, radiologists detected pneumothoraces equally well on conventional film-screen radiographs and digital images printed on film; however, they detected pneumothoraces less well on electronic viewing consoles. This latter finding reflects an important practical difference in the working behavior of radiologists interacting with a digital workstation.