Ecosystem simulation models are valuable quantitative decision tools for supporting ecosystem-based fisheries management. However, the application of ecosystem models in fisheries management is still undermined by the lack of simple procedures to test the effect of model uncertainty on policy outcomes. The use of multiple ecosystem models is viewed as “insurance” against the effects of uncertainty emanating from modelling complex systems, which calls for investigations to ascertain whether models with different structure and assumptions can give consistent policy evaluations. We compared two structurally-distinct ecosystem models, Ecopath with Ecosim (EwE) and Atlantis, for Lake Victoria by varying fishing mortality of the key functional groups: Nile perch (the top predator) and haplochromines (key prey species). We compared model behaviour at the ecosystem level and at the level of functional groups, by evaluating changes in biomass of targeted groups and the consequent effects of changes in target groups on non-target groups. Results showed qualitative similarities (direction of change) for the major harvested groups; however, the cascading effects on non-target species varied across models, depending on the species interaction feedbacks. We conclude that: EwE and Atlantis, despite the huge differences in ecological processes between the models, can give consistent qualitative advice, which is needed for strategic management decisions; consistency in the representation of trophic interactions may help to minimize variations in simulated fishery responses due to model structure. This study helps to highlight scenarios that are robust to model choice, and for which simpler models (such as EwE) could also provide reliable advice.