In this paper, we present a descriptive theory of buggy novice programs and a bug categorization scheme that is based on this theory. Central to this theory is the cognitively plausible knowledge - goals and plans - that underlies programming. The bug categorization scheme makes explicit problem-dependent goal and plan knowledge at many different levels of detail. We provide several examples of how the scheme permits us to focus on bugs in a way that facilitates generating plausible accounts of why the bugs may have arisen. In particular, our approach has led us to one explanation of why some novice programs are buggier than others. A basic part of this explanation is the notion of merged goals and merged plans in which a single integrated plan is used to achieve multiple goals.