One hundred chest radiographs, randomly selected from a hospital population, were read by 8 radiologists at 4 different levels of training and experience. All statements in the 800 interpretations were analyzed for accuracy. The analysis was based on deviation from a "true" report. Approximately 26% of significant and potentially significant statements were errors. Eighty per cent of these were false negative, 20% false positive. While significant individual differences in interpretive accuracy were found, there was no consistent pattern related to the duration of training beyond the first year of residency. Once an individual's radiologic education has progressed beyond a fundamental level, individual reader characteristics overshadow experience in the accuracy of chest film interpretation.