In the spirit of evidence-based decision making, research findings are increasingly being used to inform practice guidelines and policy making. Whether research informs the process accurately and appropriately depends on the quality of the design. This article examines the assumptions underpinning the randomised trial in relation to its application to evaluating socially complex interventions. Because the properties of the randomised trial are not independent of the characteristics of the interventions being studied, researchers need to be more attentive to selection bias, unmeasured contextual variables and uncontrolled interaction effects that arise because the environment interacts with the intervention. It is recommended that evaluations of socially complex interventions be modified by adding a complex contextual evaluation and using multiple sites.