Often a major difficulty in the design of expert systems is the process of acquiring the requisite knowledge in the form of production rules. This paper presents a novel class of neural networks which are trained in such a way that they provide an appealing solution to the problem of knowledge acquisition. The value of the network parameters, after sufficient training, are then utilized to generate production rules on the basis of preselected meaningful coordinates. Further the paper provides a mathematical framework for achieving reasonable generalization properties via an appropriate training algorithm (supervised decision-directed learning) with a structure that provides acceptable knowledge representations of the data. The concepts and methods presented in the paper are illustrated through one practical example from medical diagnosis.