In response to the increasing emphasis on 'evidence-based teaching', this article examines the privileging of randomised controlled trials and their statistical synthesis (meta-analysis). It also pays particular attention to two third-level statistical syntheses: John Hattie's "Visible learning" project and the EEF's "Teaching and learning toolkit." The article examines some of the technical shortcomings, philosophical implications and ideological effects of this approach to 'evidence', at all these three levels. At various points in the article, aspects of critical realism are referenced in order to highlight ontological and epistemological shortcomings of 'evidence-based teaching' and its implicit empiricism. Given the invocation of the medical field in this debate, it points to critiques within that field, including the need to pay attention to professional experience and clinical diagnosis in specific situations. Finally, it briefly locates the appeal to 'evidence' within a neoliberal policy framework.