This article describes an attempt to implement an expert system for production scheduling in a small manufacturing company. The approach adopted was to attempt to capture the heuristics and constraints used by the human scheduler in the form of a set of production rules that would be capable of imitating the human scheduler's behavior and eventually replace him. The implementation was eventually abandoned in favor of the introduction of a highly simplified version of the system to be used only in the decision support role, which was also unsuccessful. The reasons behind the failure are identified and discussed, and are found to relate to the following: underestimation of the complexity of the knowledge acquisition problem and the rate of change of the knowledge base; failure to adequately conceptualize the relationship between the human scheduler and the system; inadequate representation of uncertainty in scheduling parameters; the difficulty of the scheduler in adapting to an unfamiliar representation of the problem space, and lack of attention to deliberately simplifying the manufacturing operation before the introduction of the system.