A knowledge-based production system was developed for generating optimal fuel reload configurations. The system was based on a heuristic search method and implemented in Common Lisp programming language. The knowledge base embodied the reactor physics, reactor operations, and a general approach to fuel management strategy. The data base included a description of the physical system involved, i.e., the core geometry and fuel storage. The fifth cycle of the Three Mile Island Unit 1 pressurized water reactor was chosen as a test case. Application of the system to the test case revealed a self-learning process by which a relatively large number of “near-optimal” configurations were discovered. Several selected solutions were subjected to detailed analysis and demonstrated excellent performance. To summarize, applicability of the proposed heuristic search method in the domain of nuclear fuel management was proved unequivocally.