A description is presented of a performance evaluation study of three proposed distributed deadlock detection algorithms, namely, fully centralized, partially distributed, and fully distributed deadlock detection algorithms. The authors investigate and describe the extent to which these algorithms provide an increase in the level of multiprogramming, and an increase in the number of blocked, restarted transactions. A list of performance parameters and evaluation criteria is also described. It is shown that, in general, the partially distributed algorithm provides the best performance.