The trend to distributed processing is being fueled by numerous forces, including advances in technology, corporate downsizing, increasing user sophistication, and acquisitions and mergers. Increasingly, the trend in corporate information systems (IS) departments is toward sharing resources over a network of multiple types of processors, operating systems, and applications. A key design issue that arises in distributed systems concerns is how to optimally distribute data among the nodes of a network. This paper describes the problem of database distribution and proposes a methodology for the distribution of data that maximizes local references to data. The methodology considers the frequency of accesses to data from each of the nodes, and specifically accounts for the difference in cost between retrievals and updates in a distributed network. The methodology can be applied using hand calculations for smaller problems and can be also implemented in software for larger problems. (12 references) (Author/SD)