Experience from a three-year Home Office funded evaluation of a project intended to reduce school exclusions is used to explore methodological dilemmas raised by the current emphasis upon 'evidence-based' policy formation. The social construction of school exclusion rates poses problems of reliability and validity, especially when such rates are simultaneously being used for target setting. In principle, the concept of 'evidence-based' can refer to a wide variety of research questions and appropriate research methodologies. Despite this, moves towards interpreting 'evidence-based' as (...) predominantly measurement and outcomes oriented can be found both in government evaluation guidelines and in procedures for systematic reviews of research. Given the complexity of educational innovations, any neglect of research into the processes of change in naturalistic settings will not only lead to a restricted awareness of a project's impact but also to a failure to understand what certain apparent outcomes actually mean. (shrink)