A connectionist expert system (ES) in the domain of dermatology which will ultimately be used for instruction of medical students in the diagnosis of papulosquamous skin diseases is presented. The ES performs as an apparatus for diagnosis and elucidates the effects of diagnostic parameters in the derivation of viable conclusions. The ES uses the back-propagation algorithm to automatically generate a knowledge base from patient cases, each consisting of a list of symptoms as input and the expert's diagnosis as the desired output. The knowledge base has the form of a multilayered network, and knowledge is distributed over the network so that a pattern of weighted interconnections constitutes an implicit specification of decision criteria. The ES uses this knowledge base for diagnosing a patient's disease given a list of symptoms.