There are some important ambiguities in discussions about the implications of "evidence-based practice", both for educational research and for the work of teachers. In this paper we explore several of these through examining the Evidence-based Practice in Science Education (EPSE) study--part of the Teaching and Learning Research Programme (TLRP), a major UK research initiative that is explicitly committed to improving the contribution of educational research to policymaking and practice. The field of science education is in some ways a critical case, given that the notion of evidence-based practice treats scientific research as a privileged source of evidence. We examine the assumptions built into the EPSE study, and how these relate to the classical medical model of evidence-based practice, and to some other approaches to educational research that are concerned with closing the gap with practice. (Contains 14 notes.)