A universal and pressing problem at supercomputing and other research centers is the need to provide adequate management of numerous very large data sets. To be useful, these data sets will require on-line storage, random access and a comprehensive, usable cataloging scheme. This proposal presents a promising solution for a Data Management Facility (DMF) that can handle effectively the multiple uses of such data sets. The project will be a prototype development effort, and will at the same time provide a functioning facility for several scientific research communities. The latter will have greatly improved access to their data and will also work closely with us in further defining the needs of the research community.