This paper discusses a project undertaken between the Departments of Computing Science, Statistics, and the College of Veterinary Medicine to design a medical diagnostic system. On-line medical data has been collected in the hospital database system for several years. A number of induction methods are being used to extract knowledge from the data in an attempt to improve upon simple diagnostic charts used by the clinicians. They also enhance the results of classical statistical methods - finding many more significant variables. The second part of the paper describes an essentially Bayesian method of evidence combination using fuzzy events at an initial step. Results are presented and comparisons are made with other methods.