Investigates the value of applying principles derived from simulated annealing to clustering tuples in database design, and compares this technique with a graph-collapsing clustering method. It is concluded that, while the new method does give superior results, the expense involved in algorithm run time is prohibitive. (24 references) (CLB)