The term ‘interpersonal violence’ is used to refer to a wide range of violent acts, including intimate partner violence, child abuse and stranger assaults. Two extensive reviews of interventions to reduce or prevent interpersonal violence published between 1950 and 2008 identified a large literature base reflecting the extensive collective effort of violence researchers over the past 50 years. However, neither review was able to meaningfully synthesize sub-group data due to the high degree of heterogeneity present. This paper interrogates this apparent contradiction by examining three examples of predefined sub-group analyses from these reviews. None of the chosen examples produced groups of studies with adequate homogeneity for meaningful meta-analysis and synthesis, indicating that the violence research literature, while extensive, is currently too heterogeneous to be used to inform policy related to the most appropriate interventions suitable for evidence-based practice. If the literature cannot become more focused via a major topic prioritization exercise, an alternative solution to this problem may be to adopt a realist synthesis approach to determine what works, for whom, and in what context.