Distributed deadlock detection requires identifying the presence of certain properties in the global state of distributed systems. Distributed deadlock detection is complicated due to the lack of both global memory and a common physical clock, and due to unpredictable message delays. We characterize the formation and detection of distributed deadlocks in terms of the contents of local memory of distributed nodes/sites. We describe how the interaction between deadlock detection and deadlock resolution can lead to the detection of false deadlocks that are impossible to avoid due to inherent system limitations. We define shadow, phantom, and pseudo deadlocks in the proposed framework. We give examples of existing incorrect deadlock detection algorithms to illustrate how they violate the developed requirements for distributed deadlock detection. The characterization provides an insight into the properties of distributed deadlocks, expresses inherent limitations of distributed deadlock detection, and yields new correctness criteria for distributed deadlock detection algorithms. (C) 1994 Academic Press, Inc.