This article aimed to explore the realist evaluation in terms of its philosophical foundation and its methodology. It proposed the realist evaluation as a “really useful” theoretical and methodological model for education policy evaluation in the decentralization era. Firstly, three underlying assumptions of critical realism, the philosophical base of the realist evaluation, were discussed. They were stratified ontology, an open system of complexity and relative epistemology. Those philosophical hypotheses were useful as regards explaining the causation of the education policy evaluation, reflecting the complex educational fields, and accumulating the evaluation knowledge individually. Secondly, the methodology of the realist evaluation was examined. To promote the understanding of the realist evaluation methodology, Kim and Yoon(2005)’s evaluation study on Education Welfare Priority Policy was used as an illustrative case study. The basic research questions of the realist evaluation are ‘what works? for whom? in what circumstances? and how?’ To respond to this, realist evaluators formulate a programme theory, consisting of context-mechanism-outcomes. The aim of the realist evaluation is to generate, test and refine the programme theories by comparing the initial programme theories with the empirical policy evidences. This is distinguished from the mainstream education policy evaluation method: It asks ‘did it work? or not?’; its main purpose is to evaluate the intervention of an individual education policy programme. Lastly, this paper discussed the advantages and the constraints of the realist evaluation.