A model concerning the influence of implicitly activated information on cued recall and recognition is presented. The model assumes that studying a familiar word activates its associates and creates an implicit representation in long-term working memory. Test cues also activate their associates, with memory performance determined by a sampling process that operates on the intersection of information activated by the test cue with information previously activated by the studied word. Successful sampling is enhanced by preexisting connections among the associates of the studied word and by preexisting connections between it and the retrieval cue. However, the usefulness of the implicit representation is reduced by the activation of competing associates and by shifts of attention before testing. Experiments designed to test predictions of the model indicate that the associates of a familiar word can exert a powerful effect on its cued recall and recognition.