For decades there have been calls by concerned stakeholders to improve the quality of education research, and some progress has been made towards creating a more secure evidence base in some areas. More programmes and approaches that have a reasonable evidence base are now also being used in schools (but not in policy, and not necessarily because they have a reasonable evidence base). However, there has been no equivalent improvement in secure knowledge about how best to get that evidence into use, or even what difference it makes when such evidence is used. This paper looks at what little is already known about the different ways to get research evidence into use in education by summarising the results of a large-scale review of the literature. A total of 323 of the most relevant studies were looked at across all areas of public policy, and judged for quality and contribution. Very few (33) were of the appropriate design and quality needed to make robust causal claims about evidence-intouse, and even fewer of these concerned education. This means that despite over 20 years of modest improvement in research on what works in education policy and practice, the evidence on how best to deploy these findings is still very weak. We consider studies in terms of several issues, including whether they look at changes in user knowledge and behaviour, or student outcomes, and how evidence is best modified before use. Providing access to raw research evidence or even slightly simplified evidence is not generally an effective way of getting it used, even if that evidence is presented to users by knowledge-brokers, in short courses or similar. What is more likely to work for both policy and practice is the engineering of high quality evidence into a more usable format and presenting it actively or iteratively via a respected and trusted conduit, or through population measures such as legislation. Having the users actually do the research is another promising approach. Expecting each individual study they fund to have an impact is not the way forward, as this may encourage widespread use of ineffective or even harmful interventions. Publicly-funded users, including policy-makers, should be required to use evidence-led programmes from those libraries providing them and which are appropriate and relevant to their aims. Research funders should support these approaches, and help to build up libraries of successfully tested programmes. Researchers need to be scrupulous, looking at their new evidence in the context of what is already known and not looking to obtain 'impact' from single studies. More and better research is needed on the best routes for evidence-into-use. However, the improvements required of all parties are as much ethical in nature as they are technical or scientific.Keywords use of evidence, engineering of evidence, translation of evidence, research impact, knowledge transfer, robust evidence.