Ecosystem rather than species management has become an explicit part of policies that feature in international treaties and national legislation. Many of the tools that will be needed to fulfil the requirements of these policies are still in an early stage of development. One such tool is trophic ecosystem modelling. These models have been put forward to aid system-level understanding and provide insight into the potential impacts of human activities. Despite this, there are many gaps in knowledge of their strengths and weaknesses. In particular, little is known about the effect of the level of detail in a model on its performance. There has been some consideration of the effects of model formulation, as well as the effects of the physical, biological and chemical scope of multispecies and ecosystem models on their performance. A review of existing research indicates that there is a humped relationship between model detail and performance for these models, and that there are some guiding principles to consider during model development. This review gives some insight into which model structures and assumptions are likely to aid understanding and management, and which may be unnecessary. Further, it provides some understanding of whether some models can capture properties of real systems that other models cannot. The main recommendation is that the use of a single 'ultimate' ecosystem model is ill-advised, while the comparative and confirmatory use of multiple 'minimum-realistic' models is strongly recommended.