The process of roentgen diagnosis comprises three basic steps: the recording, the perception, and the interpretation of critical roentgen shadows. Volumes have been written on the recording and interpretation of these shadows, but their perception is so spontaneous that radiologists have largely taken it for granted. Errors can and do arise in this stage of roentgen diagnosis, however, and their control depends on our knowledge of the mechanisms involved. It is our purpose here, therefore, to summarize present knowledge of visual perception as it relates to roentgen diagnosis, and to suggest new research which may shed further light on this aspect. Frequency of Perceptual Error We fail to perceive significant findings in the roentgenogram with surprising frequency. With several colleagues, I investigated this problem some years ago and was amazed to find that three experienced radiologists, charged only with indicating the presence of abnormal shadows3 in a collection of chest films, “overlooked” 20 to 30 per cent of the findings which they as a group later judged to have been reportable (Table I). This sample is small, but other studies (5) indicate that expert physicians “miss” 25 to 30 per cent of “positive” chest roentgenograms under conditions in which their errors must be largely due to failures of perception (Table II) ; and military photointerpreters, in a roughly parallel study (15), failed to report an average of 54 per cent of the recorded and significant findings in the material presented to them! One cannot interpret a shadow he has not perceived, and failure of perception must, therefore, account for a substantial fraction of all our diagnostic errors. This is not to discount the importance of mistakes in judgment in the logical interpretation of perceived findings, but the control of such errors has been discussed elsewhere (8) and we shall confine this discussion to problems of perception. Let us consider, then, what is known of the physiologic mechanisms leading to perception and presumed to be responsible for this disturbing incidence of error. Search The initial phase of the process must be one of visual search (Fig. 1), and a variety of orderly search patterns have been advocated by radiologists. There has been little objective evidence, however, concerning the behavior of film readers in this respect, and Enoch (3), discussing his experimental study of the search patterns of military photointerpreters, suggests that a subjective report of the pattern followed by an observer is often misleading. Calvert and I (14), therefore, photographically recorded the search patterns of four trained radiologists in an effort to determine the sort of patterns actually employed in clinical film reading.