Abstract: The last decade has seen a number of practical decision support systems based on the normative principles of probability theory and decision theory. While, on theoretical grounds, such systems can be expected to perform well and be useful in practice, there is still little empirical data that would validate this expectation. In this talk I will present a study conducted in order to validate a normative decision support system in a laboratory setting. The study compares performance of HEPAR II, a medical system for diagnosis of liver disorders, with the performance of general practitioners on ten randomly selected patient cases with histopathologically confirmed diagnosis. The results are encouraging: HEPAR II's diagnostic accuracy was 40% higher than the best of the physicians'. The study also tests the effect of system's suggestions on the ultimate diagnosis indicated by the physicians. Here the results are encouraging as well: system's advice doubled the accuracy of the physicians.