In ecotoxicology, derivation of a "safe" environmental concentration is usually achieved by the use of extrapolation factors or by statistical extrapolation from a set of single species toxicity data. These approaches ignore ecological interactions between species in the field. An ecology-based alternative to this pragmatic approach can be ecosystem modelling, which can account for ecological interactions. However, it is largely unexplored how well the predictions of these models quantitatively agree with large-scale experimental studies. Therefore, we evaluated the capacity of a flexible ecosystem model to predict population and ecosystem-level no observed effect concentrations (NOECs) of 7 organic toxicants. These NOECs were compared with population and ecosystem-NOECs observed in 11 micro- and mesocosm studies. For each of the latter studies, the model was customized to account for the specific ecological interactions within these systems and combined with appropriate single-species toxicity data from literature. Population-NOEC predictions were accurate, or at least protective, for 60, and 85% of all considered model populations, respectively. For all 11 studies, a protective ecosystem-NOEC could be derived, being accurate in 7 cases, and conservative in 4 cases. In general, it can be stated that this type of models can serve as an ecology-based alternative to current extrapolation techniques in EEAs and water quality standard setting.