In the 21 years since social capital first appeared in the public health literature, the evidence base has grown enormously, now reaching 28 systematic reviews encompassing more than 850 individual studies. We summarise this evidence and explain why conclusions relating to both the relationship between social capital and health, and the effectiveness of interventions to promote population health remain elusive and contradictory. A critical factor is the inadequate way that context is treated in the research, and especially how context interacts with efforts to promote health in a dynamic fashion. Of all the different types of interventions one could employ to improve the health of the public, 'social capital' interventions are likely to be the most context specific and especially affected by the boundaries placed around the context. A way forward is offered that requires a combination of insights from systems thinking, community-based participatory research, and intervention and improvement sciences. This requires renewed focus on the specific components of social capital, an understanding of how context interacts dynamically with efforts to improve health, a greater role for practice in the design, implementation, adaptation and evaluation of interventions, and the support of researchers to develop better methods for recognising and classifying the knowledge generated by complex interventions.