Description-logic programs (or DL-programs for short) combine logic programs under the answer-set semantics with description logics for semantic-web reasoning. In order for a wider acceptance of the formalism among semantic-web engineers, it is vital to have adequate tools supporting the program development process. In particular, methods for debugging DL-programs are needed. In this paper, we introduce a framework for interactive stepping through a DL-program as a means for debugging which builds on recent results on stepping for standard answer-set programs. To this end, we provide a computation model for DL-programs using states based on the rules that a user considers as active in the program and the resulting intermediate interpretation. During the course of stepping, the interpretations of the subsequent states evolve towards an answer set of the overall program. Compared to the case of standard answer-set programs, we need more involved notions of states and computations in the presence of DL-atoms. In particular, if non-convex DL-atoms are involved, we have to allow for non-stable computations. Intuitively speaking, we realise this by allowing the user to assume the truth of propositional atoms which must be justified in subsequent states. To keep track of these additional atoms, we extend the well-known notion of an unfounded set for DL-programs.