Abstract. Description Logic programs (DL-programs) are a prominent approach for a loose coupling of rules and ontologies, which has become a topic of increased interest. When computing answer sets of a DL-program, special DL-atoms, which provide query interface to an ontology, are evaluated under a possibly changing input that gives a context for the evaluation. Many different such contexts may exist and evaluating a DL-atom may be costly even for one context. Thus a natural question to ask is when the evaluation is independent of the context. Such information has immediate applications in optimization of DL-programs, but is also beneficial for other reasoning tasks, like inconsistency diagnosis and program repair. We provide an answer to this question based on a semantic notion of independence and provide a complete characterization of independent DL-atoms. We then extend the characterization to independence under additional information about inclusions among rule predicates. Moreover, we develop an axiomatization which allows one to derive all tautological DL-atoms in the general case and under predicate inclusions. A complexity analysis reveals that checking whether a DL-atom is independent, can be done efficiently.