The testing effect is the finding that completing a test rather than restudying is beneficial for memory performance. One explanation of the testing effect is the mediator effectiveness hypothesis. This hypothesis suggests while completing a test, a mediator (M) will become activated and enhance the link between a cue (C) and a target (T). Two experiments provided evidence for limitations of the mediator effectiveness hypothesis. In Experiment 1, semantic mediator cues were not more effective as cues for target recall compared to target-related cues. As well, participants were more likely to make an original cue error when presented with semantic mediator cues than target-related cues. In Experiment 2, shallow processing during encoding impaired the effectiveness of semantic mediator cues. These experiments identify limitations of the mediator effectiveness hypothesis as a general explanation of the testing effect. Alternative explanations of the testing effect should address the role of backward associations from the mediator to the original cue and the role of depth of encoding during original learning.