The current method for assessing long-term risk of pesticides to mammals in the EU is based on the individual rather than the population-level and lacks ecological realism. Hence there is little possibility for regulatory authorities to increase ecological realism and understanding of risks at the population-level. Here we demonstrate how, using ABM modelling, assessments at the population-level can be obtained even for a pesticide with complex long-term effects such as epigenetic transmission of reproductive depression. By objectively fitting nonlinear models to the simulation outputs it was possible to compare population depression and recovery rates for a range of scenarios in which toxicity and exposure factors were varied. The system was differentially sensitive to the various factors, but vole ecology and behaviour were at least as important predictors of population-level effects as toxicology. This emphasises the need for greater focus on animal ecology in risk assessments.