Population-level models are increasingly recognised as potentially powerful tools in environmental risk assessment. However, the practical application of dynamic population predictions for decision making is not straight forward in everyday risk assessment of down-the-drain chemicals. Probabilistic environmental risk assessment (PERA) has been suggested as a way to make uncertainty more explicit and to account for spatial and temporal variability. The outcome of a PERA is typically a measure of expected risk with an associated uncertainty interval. We present a conceptual framework that explores how probabilistic approaches can be applied in population models to incorporate more ecological relevance into risk assessments while keeping both uncertainty and variability explicit and transparent. One of the key challenges is understanding the protection goals for ecological scenarios exposed to anthropogenic stressors. For instance, systems that are already impaired by high volume emissions of untreated wastewater may require modified protection goals (e.g. protection of microbial purification processes and recovery of food web structure and diversity). For higher organisms, the protection goals must be reflected by a defined set of endpoint metrics that can quantify changes in population-level dynamics. These metrics must be carefully selected based on both the specific scenario and protection goal, as different options will lead to very different interpretations of effect. Finally we discuss the importance of making the relation between the willingness to accept risk and the severity of the effect explicit to facilitate decision-making. We consider these discussions a necessary first step in bringing the full potential of population-level models into risk assessment of down-the-drain chemicals.