The complexity of large-scale systems can be controlled by imposing structure on the problem space. In the domain of device diagnosis the problem space consists of all possible faults and their manifestations which cannot be represented directly. A multitude of perspectives, such as physical structure, function, and behavior, need to be modeled to describe this space implicitly. The utility of a multilayered knowledge representation scheme for a relation-based knowledge base is demonstrated. The structure established by multiple layers and multiple perspectives facilitates classification of symptoms and focuses on the exploration of the solution space. >