During the last decade there has been a wealth of papers discussing the development of scoring systems in medicine, some of which have led to fully operational computer-aided diagnostic systems. In this paper we sketch the development of a simple scoring system for one of the more common rheumatic diseases, rheumatoid arthritis (RA) using a two-tier model, independence Bayes' followed by logistic discrimination. The scoring system gives reasonably well calibrated probability estimates of RA which suggests ways in which the development of computerised systems in rheumatology might be approached.