In this paper we investigate the robustness of several deadlock detection algorithms for distributed computing systems. We analyze the behavior of each algorithm in the presence of two classes of failures - lost messages and single site failures. In the case of single site failure we consider six different types of sites depending on how they can participate in deadlock and deadlock detection. The observation and conclusions made in this paper are intended to show how robust the present algorithms are and to provide an insight and better understanding of distributed algorithms robustness. (Author)