Language processing necessarily relies on memory of the immediate past; previously encoded information needs to be retrieved to incorporate new information successfully and efficiently. One of the areas in which the role of memory retrieval is prominent is referential processing where one or more referential candidates are initially encoded and then are subsequently retrieved when a referring expression (such as a pronoun) is encountered. In this talk, I will report the results of three projects investigating whether and how retrieval difficulty affects referential processing. In all projects, we manipulated retrieval difficulty by varying the amount of extra information attached to potential referents, producing representationally rich (e.g., the actor who had recently won an Oscar award) and bare (e.g., the actor) referential candidates. We measured behavioral responses such as the choice between different forms of referring expressions (e.g., pronouns vs. repeated nouns) during language production, looking probability in the Visual World as well as Event Related Potentials, and observed that representational richness facilitates the retrieval of associated memory items. Taken together, the results lend support to cue-based retrieval theories of language processing according to which richer memory representations should be more easily retrievable due to reactivation and/or distinctiveness.