‘Mechanism’ is crucial to realist evaluation. Yet it has turned out to be tricky. The term is used in many different ways, not all of which accord with Pawson and Tilley’s conception of realist evaluation. In this talk, I’ll return to the original use of the term and go on to illustrate ways in which it has been applied in practice in ways that do and do not accord with Pawson and Tilley. I’ll say a little about developing and testing realist hypotheses in evaluative studies. I’ll conclude more speculatively on the contribution realist evaluations of programmes and policies might make to cumulation in social science.