A set of 8 symptoms was asked of a group of 204 patients. A computer was loaded with the answers to these questions, and also the correct radiologic diagnosis of each patient. Then, the same set of symptoms was asked of another group of 96 patients and loaded in the computer. The primed computer was then asked to predict the radiographic diagnoses of this unknown group of patients. The computer correctly chose 73% of hiatal hernia patients, 69% of duodenal ulcer patients, 27% of gastric ulcer patients, 75% of gallstone patients, 38% of functional patients, and 33% of gastric carcinoma patients. The basic limitation to high diagnostic accuracy using subjective symptoms is the variability of the data obtained from the patients. This variability was related to the inherent causes of symptoms, as well as the patient''s limitations as a witness, and the physician''s limitations in extracting historical data. Symptom diagnosis may be improved by weighing symptoms differently, and altering our attitude toward the date collected during an interview. This kind of computer study offers good possibilities for assisting in analysis of symptoms by the office physician and improving the teaching of medical residents.