A wealth of literature suggests individuals use feelings in addition to facts as sources of information for judgment. This paper focuses on a manipulation in which participants list either a few or many examples of a given type, and then make a judgment. Instead of using the number of arguments or evidence strength, participants are hypothesized to use the subjective ease of generating examples as the primary input to judgment. This result is commonly called the ease-of-retrieval effect, and the feeling of ease is typically assumed to mediate the effect. We use meta-analytic methods across 142 papers, 263 studies, and 582 effect sizes to assess the robustness of the ease-of-retrieval effect, and whether or not the effect is mediated by subjective ease. On average, the standard few/many manipulation exhibits a medium-sized effect. In experimental conditions designed to replicate the standard effect, about one third to one half of the total effect is mediated by subjective ease. This supports the standard explanation, but suggests that other mediators are present. Further, we find evidence of publication bias that reduces the standard effect by up to one-third. We also find that (1) moderator manipulations that differ from the standard manipulation lead to smaller, often reversed effects that are not as strongly mediated as ease, (2) several manipulations of theory-based moderators (e.g., polarized attitudes, misattribution) yield strong theory-consistent effects, (2) method-based moderators have little or no effects on the results, and (4) the mediation results are robust with respect to assumptions about error structure.