A methodology is presented for creating models which are suitable for use in fault detection and diagnosis schernes in applications where it is impossible to obtain data from the actual plant. Generic qualitative models based on fuzzy rules are used to describe the basic features of the behaviour of a class of plants of similar design. The generic models are identified off-line from training data produced by computer simulation of typical plant designs. The rnethod uses a data clustering algorithm to assist in identifying structure and a fuzzy identification algorithm to estimate the parameters of the model from the training data. A measure of the degree of similarity between fuzzy models is introduced to determine the extent to which the behaviour of individual plants is similar to that of the generic model.The method is used to generate generic models of a cooling coil subsystem when it is operating correctly and when the coil has become fouled. Results are presented which show that the generic models can be used to describe a, class of plant designs without greatly increasing the ambiguity associated with the fault diagnosis.