This article considers different approaches to policy evaluation within recent writings on governance and evaluation research. Governance theorists locate evaluation within a new ‘mode of regulation’ aimed at managing a ‘dispersed state at arms length’. The strength of this approach is that it places policy evaluation within an explicitly political framework. However, the governance perspective focuses on evaluation as a state-centred activity and tends to underplay its contested nature. Within the evaluation research literature, on the other hand, attempts to deal with the ‘politics of evaluation’ tend to focus on evaluation as a research practice. This misses the discursive construction of evaluation as a political project. While some approaches within the evaluation literature stress the ‘democratising’ role of ‘stakeholder participation’, they are set within a democratic pluralist model of politics, which frequently fails to deal with underlying power relations within which the clash of competing stakeholder interests takes place. This article challenges these constructions. It draws on the strengths of both approaches to propose a critical ‘politics of evaluation’ based on service user control. It suggests some practical strategies drawn from a review of recent evaluation research.