We describe the architecture and query-answering algorithms used in the Information Manifold, an implemented information gathering system that provides uniform access to structured information sources on the WorldWide Web. Our architecture provides an expressive language for describing information sources, which makes it easy to add new sources and to model the ne-grained distinctions between their contents. The query-answering algorithm guarantees that the descriptions of the sources are exploited to access only sources that are relevant to a given query. Accessing only relevant sources is crucial to scale up such a system to large numbers of sources. In addition, our algorithm can exploit run-time information to further prune information sources and to reduce the cost of query planning.