Findings on plain chest radiographs of patients with aortic dissection are variable and often overlap those of patients without dissection. To determine which findings were most useful in predicting aortic dissection, plain chest radiographs from 36 patients with aortographically proven aortic dissection and 36 patients from a control population were randomized and analyzed independently by five radiologists for the presence of various radiographic features associated with this condition. A widened aortic knob, widened descending aorta, and widened mediastinum showed the greatest interobserver agreement (p less than .001) although the overall interobserver agreement was poor. The final conclusion of the radiologists was a better predictor of dissection than any of the individual radiographic features alone. Widening of the mediastinum (p less than .001) and widening of the aortic knob (p less than .012) were the only two radiographic features of significance in predicting dissection. In a stepwise multiple logistic regression model, the radiologists achieved an overall accuracy of 85%, a sensitivity of 81%, and a specificity of 89%. Although this illustrates the usefulness of plain chest radiographs in diagnosing aortic dissection, poor interobserver agreement dictates that further definitive investigation be undertaken.