The modality effect is a central issue in multimedia learning [see Mayer (Cambridge University Press, 2005a), for a review]. Sweller's Cognitive Load Theory (CLT), for example, presumes that an illustrated text is better understood when presented visually rather than orally. The predictive power of CLT lies in how it links in to Baddeley's (1986) model of working memory and Penney's ("Mem Cognit" 17:398-442, 1989) Separate-Streams Hypothesis. Ginns's ("Learn Instr" 4:313-331, 2005) recent meta-analysis also supports the modality effect (d = 0.72, based on 43 independent effects). This article replicates the meta-analysis of the modality effect based on 86 independent effects (with "within-study subgroups" as the unit of analysis and with "mean of the outcomes" as the dependent measure), with results showing a reduction of the overall effect size by almost half (d = 0.38), and even more when Duval and Tweedie's "Trim and Fill" method is used to correct publication bias (d = 0.20). This article also widens the scope of the analysis of moderator variables (e.g. Pace of presentation, Type of visualization, Research group) as well as their potentially confounded effects. Finally, it is argued that, for theoretical reasons, the so-called modality effect cannot be based on Penney's or Baddeley's theories and must be explained in a different way.