We report on our experience in developing a query-answering system that integrates multiple knowledge sources. The system is based on a novel architecture for combining knowledge sources in which the sources can produce new subgoals as well as ground facts in the search for answers to existing subgoals. The system uses a query planner that takes into account different query-processing capabilities of individual sources and augments them gracefully. A reusable ontology provides a mediated schema that serves as the basis for integration. We have evaluated the system on a suite of test queries in a realistic application to verify the practicality of our approach.