The elaborative retrieval account of retrieval-based learning proposes that retrieval enhances retention because the retrieval process produces the generation of semantic mediators that link cues to target information. We tested 2 assumptions that form the basis of this account: that semantic mediators are more likely to be generated during retrieval than during restudy and that the generation of mediators facilitates later recall of targets. Although these assumptions are often discussed in the context of retrieval processes, we noted that there was little prior empirical evidence to support either assumption. We conducted a series of experiments to measure the generation of mediators during retrieval and restudy and to examine the effect of the generation of mediators on later target recall. Across 7 experiments, we found that the generation of mediators was not more likely during retrieval (and may be more likely during restudy), and that the activation of mediators was unrelated to subsequent free recall of targets and was negatively related to cued recall of targets. The results pose challenges for both assumptions of the elaborative retrieval account.