Various strategies can be used in the diagnosis of upper gastrointestinal tract bleeding. This study investigates the relevance of anamnestic and clinical findings for the diagnosis of the bleeding source. The authors introduced a computer-aided diagnostic system using Bayes' theorem and compared it with clinicians' predictions using anamnestic and clinical findings only. There was no difference in the overall accuracy rates, but a difference was observed in the diagnostic behaviors of the two "systems." In addition, the discriminatory ability of the computer-aided system, the sharpness of the predictions obtained, and the reliability of the posterior probabilities were analyzed. It is concluded that the clinician and the computer-aided system are not able to discriminate well between the disease categories. Derived classification matrices and probability-based measures show the reasons for the inadequacy of diagnostic information obtainable from the clinical history and physical findings.