Evidence syntheses are used to inform health care policy and practice. Behaviour change theories offer frameworks for categorising and evaluating interventions and identifying likely mechanisms through which effects are achieved. Yet systematic reviews rarely explicitly classify intervention components using theory, which may result in evidence syntheses and health care practice recommendations that are less than optimal. This paper outlines a method for applying theory to evidence syntheses of behaviour change interventions. We illustrate this method with an analysis of 'audit and feedback' interventions, based on data from a Cochrane review. Our analysis is based on Control Theory, which suggests that behaviour change is most likely if feedback is accompanied by comparison with a behavioural target and by action plans, and we coded interventions for these three techniques. Multivariate meta-regression was performed on 85 comparisons from 61 studies. However, few interventions incorporated targets or action plans, and so meta-regression models were likely to be underfitted due to insufficient power. The utility of our approach could not be tested via our analysis because of the limited nature of the audit and feedback interventions. However, we show that conceptualising and categorising interventions using behaviour change theory can reveal the theoretical coherence of interventions and so point towards improvements in intervention design, evaluation and synthesis. The results demonstrate that a theory-based approach to evidence synthesis is feasible, and can prove beneficial in understanding intervention design, even where there is insufficient empirical evidence to reliably synthesise effects of specific intervention components.