The use of modelling approaches in marine science, and in particular fisheries science, is explored. We highlight that the choice of model used for an analysis should account for the question being posed or the context of the management problem. We examine a model-classification scheme based on Richard Levins' 1966 work suggesting that models can only achieve two of three desirable model attributes: realism, precision, and generality. Model creation, therefore, requires trading-off of one of these attributes in favour of the other two: however, this is often in conflict with the desires of end-users (i.e. mangers or policy developers). The combination of attributes leads to models that are considered to have empirical, mechanistic, or analytical characteristics, but not a combination of them. In fisheries science, many examples can be found of models with these characteristics. However, we suggest that models or techniques are often employed without consideration of their limitations, such as projecting into unknown space without generalism, or fitting empirical models and inferring causality. We suggest that the idea of trade-offs and limitations in modelling be considered as an essential first step in assessing the utility of a model in the context of knowledge for decision-making in management.