Quantitative ecosystem indicators are needed to fulfill the mandate for ecosystem-based fisheries management. A variety of community metrics could potentially be used, but before reference levels for such indices can be established the sensitivity of candidate indices to fishing and other disturbances must be determined. One approach for obtaining such information is to test candidate indicators with models that mimic real ecosystems and can be manipulated experimentally. Here we construct a size-based multi species model of a community of fish species that interact by predation. The model was parameterized for 21 fish species to obtain a predation-regulated community. Following an analysis of the sensitivity of the model to parameter uncertainty, we tested the sensitivity of community-level indicators to increasing levels of fish mortality (F). Abundance and biomass spectra were sensitive to fishing mortality, with the slope decreasing with increasing F. Species diversity size spectra were also very sensitive to F; with diversity in the largest size classes declining rapidly. In contrast, k-dominance curves were less sensitive to fishing pressure. Importantly, however, although most community-level metrics showed clear trends in response to fishing, single-species declines in spawning stock biomass were the most sensitive indicators of fishing effects.