Scientists and intelligence analysts are interested in quickly discovering new results from the vast amount of available geospatial data. The key issues that arise in this pursuit are how to cope with new and changing information and how to manage the steadily increasing amount of available data. This paper describes a novel agent architecture that has been developed and tested to address these issues by combining innovative approaches from three distinct research areas: software agents, georeferenced data modeling, and content-based image retrieval (CBIR). The overall system architecture is based on a multi-agent paradigm where agents autonomously search for images over the Internet, then convert the images to a vector used for use in searching and retrieval. Results show that this system is capable of significantly reducing the time and management effort associated with large amounts of image data.