The dynamics of fish stocks are greatly influenced by predatory and fishery removals. We describe the strengths and weaknesses of a suite of approaches that assess the scope and magnitude of fishery and predatory mortality in aquatic ecosystems. We examine: (i) empirical models for pattern detection, process exploration, and model parameterization (e.g. multispecies time series and multiple regression); (ii) qualitative process-based models of species interactions (e.g. food web analyses and loop analyses); and quantitative process-based multispecies models that include: (iii) static flow models (e.g. ECOPATH); (iv) dynamic models of either numerical abundance (e.g. Lotka-Volterra) or biomass (e.g, ECOSIM and multispecies biomass dynamics); and (v) spatially explicit versions of the above. Decisions concerning a specific modelling approach should be evaluated in the context of the goals of the model and empirical database. Important considerations include model use, desired output, and management context. Four important factors to consider in model selection are the spatial and temporal extent and resolution, conservation of mass or numbers, mathematical representation of predator-prey interactions, and mathematical representation of technical interactions. We conclude that the most effective modelling exercise to achieve a given set of goals should incorporate the evaluation of multiple model configurations.