This research project is an experiment in the acquisition, use, and revision of knowledge for expert systems. The system, named PROTOS, interacts with a human expert to elicit knowledge. PROTOS then independently applies this knowledge to perform the expert task. Our research addresses three fundamental issues in concept formation and classification: First, PROTOS learns and uses naturally occurring, partially undefined concepts. Second, PROTOS uses learned knowledge in multiple ways, as people do, to generate examples of a concept, to predict unseen features of an object, to interpret ambiguous examples, and to explain its classification decisions. Third, PROTOS learns from teacher-supplied explanations, so that human experts can facilitate knowledge transfer beyond simply giving examples. Computer systems with intelligent behaviour must acquire and reason with formidable amounts of knowledge. Because it improves the methods by which computers can absorb knowledge, this research is vitally important. The PROTOS system itself will first be applied, as a testbed example, to the task of learning expert knowledge for clinical audiology.