This paper explores some of the assumptions underlying ‘evidence based’ approaches to poverty reduction impact assessment. It argues that the discourse of Evidence-Based Policy (EBP) offers poor guidance to those who seek to ensure that social policy making is informed by the findings of social science. EBP discourse relies on a technocratic, linear understanding of the policy making process and on a naïve empiricist understanding of the role of evidence. This renders it unable to engage with the role of the underlying discursive frameworks and paradigms that render evidence meaningful and invest it with consequence: EBP discourse does not help us understand either how policy changes, or what is at stake in dialogue across the ‘research-policy divide’. Rather than simply focusing on evidence, approaches to policy change need to focus on how evidence is used in the politically loaded and ideologically compelling ‘policy narratives’ that contest rival policy frameworks. The paper considers an example from the South African context – the shift to the ‘two economies’ framework and the policy interventions associated with ASGISA – and explores the implications for approaches to research more attuned to the realities of the policy making process. It concludes with a discussion of the implications for social researchers and policy makers.