Frame-based knowledge representation and reasoning systems typically provide procedural interfaces for asking about properties of individuals and concepts. We propose an alternative declarative approach that extends standard interface functionality by supporting selective viewing of components of complex objects. Instead of just returning sets of individuals, our queries match concepts and filtered fragments of descriptions. The query language is an extended form of the language used to describe the knowledge-base contents, thus facilitating user training. In this paper, we describe a variety of possible semantics for answering queries in description logics. We investigate the algorithms required when answers are deduced by matching queries against a \structural normal form" of descriptions. As part of our approach, we introduce a useful refinement of the notion of structural subsumption.