Troubleshooting is a common form of problem solving. Technicians (e.g., automotive mechanics, electricians) and professionals (physician, therapists, ombudspersons) diagnose faulty systems and take direct, corrective action to eliminate any faults in order to return the systems to their normal states. Traditional approaches to troubleshooting instruction have emphasized either theoretical or domain knowledge about the system or specific troubleshooting procedures. These methods have failed to develop transferable troubleshooting skills in learners. In this article, we propose an architecture for designing learning environments for troubleshooting. The architecture integrates experiential, domain, and device knowledge in a learning system that enables learners to generate and test hypotheses for every action they take, relate every action to a conceptual model of the system, and query experienced troubleshooters about what they would do. The architecture includes three essential components: A multi-layered conceptual model of the system that includes topographic, function, strategic, and procedural representations; a simulator that requires the learner to generate hypotheses, reconcile the hypotheses to the system mode, test the hypotheses, and interpret the results from the test; and a case library that uses a case-based reasoning engine to access relevant stories of troubleshooting experiences as advice for the learner. This novel architecture can be used to develop learning environments for different kinds of troubleshooting.