Robust projections of future trends in global fish biomass, production and catches under different fishing scenarios are needed to inform fisheries policy in a changing climate. Trust in future projections, however, relies on establishing that the models used can accurately simulate past relationships between exploitation rates, catches and ecosystem states. Here we use fisheries catch and catch-only assessment models in combination with effort data to estimate regional fishing exploitation levels (defined as the fishing mortality relative to fishing mortality at maximum sustainable yield, F/FMSY). These estimates are given for large pelagic, forage and demersal fish types across all large marine ecosystems and the high seas between 1961 and 2004; and with a ‘ramp-up’ between 1841-1960. We find that global exploitation rates for both large pelagic and demersal fish are consistently higher than for forage fish and reached their peaks in the late 1980s. We use the exploitation rates to globally simulate historical fishing patterns in a mechanistic fish community model. We find a good match between model and reconstructed fisheries catch, both for total catch as well as catch distribution by functional type. Simulations show a clear deviation from an unfished model state, with a 25% reduction in fish biomass in large pelagic and demersal fish in shelf regions in the most recent years and a 50% increase in forage fish, primarily due to the release of predation pressure. These results can set a baseline from which the effect of climate change relative to fishing could be estimated.