In this paper, we incorporate the notion of rough sets into terminological logics. Terminological logics formalize the classical frame-based knowledge representation systems in AI and can represent and reason about concepts and roles in the objective worlds. In such logics, a concept is interpreted as a class of individuals, while a role is a binary relations between them. Since the extensions of concepts have rigid boundaries, the systems can not handle rough concepts. By integrating rough set theory with terminological logics, we can model rough concepts and their approximations and reason about the rough subsumption between concepts in the systems. In our framework, two individuals are discernible with respect to a role if they have diierent relationship with any individual in this role. Thus a variety of indiscernibility relations can be determined and we can represent and reason about data of diierent granularities in a common language.