This paper deals with a method for diagnosing by using the so-called descriptive knowledge bases. It is based on the concept of using a MIMO fuzzy model to represent the information contained in the knowledge base. Model structure and initial parameter values are estimated by applying the FMEA. Finer parameter estimates are obtained by using a backpropagation learning algorithm. Rules for learning parameters of a MIMO fuzzy model are derived. Actual diagnostic problem is solved by using an an optimization algorithm for inverting the MIMO fuzzy model.