Program modules and data structures are interconnected by calls and references in software systems. Partitioning these entities into clusters reduces complexity. For very large systems manual clustering is impractical. A method to perform automatic clustering is described and a metric to quantify the complexity of the resulting partition is developed. © 1981.