In this paper we investigate two major topics; firstly, througha survey given to graduate students in a parallel message passing programming class, we categorize the errors they made (and the ways they fixed the bugs) into a number of categories. Secondly,we analyze these answers and providesome insight into how software could be built to aid the development, deployment, and debugging of parallel message passing systems. We draw parallels to similar studies done for sequential programming, and finally show how the idea of multilevel debugging relates to the results from the survey.