We present a general preferential semantic framework for plausible subsumption in description logics, analogous to the KLM preferential semantics for propositional entailment. We introduce the notion of ordered interpretations for description logics, and use it to define two mutually dual non-deductive subsumption relations. We outline their properties and explain how they may be used for inductive and abductive reasoning respectively. We show that the preferential semantics for subsumption can be reduced to standard semantics of a sufficiently expressive description logic. This has the advantage that standard DL algorithms can be extended to reason about our notions of plausible subsumption.