Good practice in primary research has evolved over many decades of research in applied linguistics to counter human fallibility and biases. Surprisingly, perhaps, synthesising such research in an entire field has only recently started to develop its own methodologies and recommendations. This paper outlines some of the issues involved, especially in terms of quantitative research and meta-analysis. A second-order synthesis of meta-analyses in Computer-Assisted Language Learning (CALL) provides only medium effect sizes, but the figures are interpreted in terms of realistic expectations. The inevitable variation in effect sizes can be attributed in principle either to the research methodologies (both primary and secondary) or – more interestingly – to real-world phenomena.