In the past, the medical diagnostician has performed abstraction and generalization of observations subconsciously without formulation of the process. This made the procedure a highly individual art requiring knowledge, skill, and experience. Consequently, a general description of diseases sufficed, without precise definition of distribution and variance of symptoms. With the newly invented logical machines, however, algebraization of the diagnostic process or its simulation by mathematical models has become inevitable. Lusted and Ledley (6) will always have the credit for being the first to introduce an applicable model of diagnostic evaluation. In spite of criticism, the Bayes theorem made general studies possible. From these, good and reproducible results were achieved, especially after Warner (20) formulated concrete definitions of applicability. Further improvement may be expected with application of Brunk and Lehr's theoretical concept (1). In this study we sought to examine the applicability of a program, based on the Bayes theorem, in a population differing in culture and medical tradition from those of the original. The underlying definitions of diseases will be compared and discussed. Winkler, Kloss, and I applied the thyroid program written by Fitzgerald and Williams (2, 3, 21, 22) and based on the studies of Williams and Overall (7–10, 21) to patients of the Department of Nuclear Medicine of the University in Bonn, Germany. The results, reported elsewhere in detail (12–14, 23), are summarized below. Ninety-three correct classifications of thyroid function were possible in 100 cases of hyperthyroidism. This agreement was increased to 95 per cent with use of a program updating the underlying incidence matrices. The program is based on 17 clinical signs and symptoms and 5 laboratory tests. Of 105 nonselected sequential cases with varying diagnoses, the computer diagnosis was the same in 96 cases without laboratory data as the clinical diagnosis based on multiple laboratory tests and clinical observations. The diagnoses were made by an experienced staff (4). These results are remarkable when the possibility of differences in base rates, distribution of symptoms, and laboratory ranges is considered (17). The matrix and base rates of Fitzgerald and Williams' study were originally based on 879 cases and later enlarged by 500 cases (2). Table I compares their base rates for the three possible diagnoses of thyroid function to the base rates after our 205 cases were added to the data of the original matrix. Only hypothyroidism shows the same tendency as in the enlarged matrix of Fitzgerald and Williams. Hyperthyroidism is favored by the experimental design, however, since in the first series 100 selected hyperthyroid cases had been processed. Such a selection should be avoided when the examination of the base rates alone is intended.