Current methodologies for information integration are inadequate for solving the problem of integration of large scale, distributed information sources (e.g. databases, free-form text, simulation etc.). The existing approaches are either too restrictive and complicated as in the “federated” (global model) approach or do not provide the necessary functionality as in the “multidatabase” approach. We propose a hybrid approach combining the advantages of both the federated and multidatabase techniques which we believe provide the most feasible avenue for large scale integration. Under our architecture, the individual data site administrators provide anaugmented export schemaspecifying knowledge about the sources of data (where data exists), their structure (underlying data model or file structure), their content (what data exists), and their relationships (how the data relates to other information in its domain). The augmented export schema from each information source provides an intelligent agent, called the “mediator”, knowledge which can be used to infer information on some of the existing inter-system relationships. This knowledge can then be used to generate a partially integrated, global view of the data.