In 2 main experiments, the mediated priming effect was used to determine whether retrieval continues to be mediated after repeated testing. In each experiment, participants used the keyword method to learn French vocabulary, then completed a modified lexical decision task in which they first translated a French word, and then made a lexical decision on a semantic associate either of the keyword or of the English translation, a semantically unrelated word, or a nonword. In Experiment 2, the amount of testing was varied before the lexical decision task. A mediated priming effect was found with limited testing, consistent with mediated retrieval as proposed by both the working memory mediation model and the mediator effectiveness hypothesis, but not after extended testing, consistent with direct retrieval as proposed by the direct access model. (Contains 3 figures.)