The paper describes a model based fault diagnosis scheme which uses explicit fuzzy reference models to describe the symptoms of both faulty and fault-free plant operation. The reference models are generated from training data which are produced by computer simulation of typical plant. A fuzzy matching scheme compares the parameters of a fuzzy partial model, identified online using normal operating data collected from the real plant, with the parameters of the reference models. The reference models are also compared to each other to take account of the ambiguity which arises at some operating points when the symptoms of correct and faulty operation are similar. Basic assignments, which indicate the strength of the evidence that the system is operating correctly or has a particular fault, are calculated from the fuzzy measures of the similarity. Results are presented which demonstrate the applicability of the scheme.