One of the most important and challenging knowledge management problems faced by NASA is the integration of heterogeneous information sources. We have designed and implemented a generalized data mediation architecture called SemanticIntegrator to address this problem. In contrast with specialized integration solutions, this architecture is more easily reused for different domains and information sources, resulting in reduced software engineering costs compared to conventional hard-coded solutions. Our approach uses semantic integration techniques to combine information sources based on semantic models of the stored data and explicit integration rules. The architecture uses individual data source ontologies to wrap each data source plus an integrating ontology that serves as the user's task-specific view of the integrated data. Rules translate data between the data source and integrating ontologies. As a first test of the SemanticIntegrator architecture, we built a system to demonstrate information integration in the context of planetary exploration operations. Although semantic integration technologies show great promise for addressing NASA’s complex information management needs, more work is necessary to design and implement mapping approaches that are sophisticated, yet understandable and maintainable for real-world applications.