Anthropogenic chemicals are essential for modern society, but many of these man-made chemicals enter the environment one way or another. For the last 10 years, the chemical industry in Europe has doubled in production and is expected to double again in the next 10 years. Ecological risk assessment aims to estimate the concentration of harmful substances in the environment and the associated effects on the ecosystem. Based on the estimated concentrations and the predicted effects, potential risks for adverse effects to the environment are identified. Effect assessment nowadays is performed using standardized toxicity tests, with individual organisms, exposed to single substances, in strictly controlled laboratory conditions. However, the environment is complex, as we have individuals living together in populations, exposed to mixtures of chemical substances, under varying environmental conditions. Mechanistic effect models have gained increasing interest from the scientific community, as they can extrapolate effects across biological levels, i.e., from sub-organismal to the population or community level. Yet, applications of mechanistic effect models for mixture toxicity in a population context are limited. The aim of the current thesis is to demonstrate the use of mechanistic models to predict population-level effects of mixtures. Focus is on the freshwater crustacean Daphnia magna (the water flea), two metals (copper and zinc), and four organic priority substances listed under the Water Framework Directive (pyrene, dicofol, alfa-hexachlorocyclohexane, and endosulfan). A generic individual-based model (IBM) implementation of the dynamic energy budget (DEB) theory was used to predict mixture toxicity effects to D. magna populations. Toxic stress was predicted using DEB-TKTD (extension of DEB including toxicokinetic-toxicodynamic processes) for sub-lethal effects and GUTS-RED-SD (reduced version of the general unified threshold model for survival assuming stochastic death) for lethal effects. A mixture toxicity implementation was developed, based on the general statistical models for mixture toxicity used in risk assessment. Two mixture toxicity approaches in mechanistic effect models can be considered: independent action or damage addition. In a first case (Chapter 2), we extrapolate effects observed at the individual level to relevant population-level effects of mixtures. The model was applied for mixtures of copper and zinc. The DEB- TKTD and GUTS sub-models were calibrated based on data from a standard 21-day chronic reproduction test (endpoints: growth, reproduction, and survival over time). A population experiment with mixtures of copper and zinc was performed. The DEB-IBM, assuming independent action for mixture toxicity, was able to reproduce the effects observed in the population experiment. Using the DEB-IBM, the observed trends were explained. The absence of zinc effects was explained through population-level compensation mechanisms. The increased mortality due to zinc is compensated by a decrease in starvation-related mortality. For copper, the switch from copper-induced mortality to starvation-related mortality explained the recovery over time observed in the experiment. Based on standard toxicity data at the individual level, mixture toxicity effects at the population were predicted. Based on the DEB-TKTD theoretical model, we hypothesize that combinations of physiological modes of action (PMoAs) in DEB-TKTD can lead to diverging effects at the population level. As a matter of fact, the PMoA will determine how the energy from food is redistributed within the population under chemical stress. We used DEB-IBM to design a population experiment, testing specific combinations of substances based on their inferred PMoAs (Chapter 3). We tested combinations of four organic substances: pyrene, dicofol, alfa-hexachlorocyclohexane (α-HCH), and endosulfan. An independent validation of mixture toxicity effects at the population level was performed with blind predictions, calibrated on individual-level effects of single substances only. Strong correlation was found between data and predictions during the constant exposed phase, the recovery phase after, and the pulsed acute phase. However, the recovery after the acute phase was not well predicted, meaning the model is unreliable in situations with high lethality. Overall, the independent action approach correctly predicted the observed mixture effects in the population experiment. The damage addition model was tested for the HCH-endosulfan mixture, but overpredicted the effects. Interestingly, synergisms (compared to statistical independent action) were observed in the population experiment that were correctly predicted by the DEB-IBM. We initially hypothesized that increased or decreased effects can occur due to the linking of DEB energy flows within the population. Overall, DEB-IBM was better in predicting mixture toxicity at the population level than current statistical models used in risk assessment. The two cases have shown the validity and relevance of mechanistic population models for mixture toxicity risk assessment. Application of these models for regulatory risk assessment is currently limited. We envision applications of mechanistic population models in current European regulations that encompass the risk assessment of chemicals, such as REACH, PPP, and BPR (Chapter 4). In this chapter, three example applications are highlighted. In a first example, mechanistic population models are used as predictive tools for the risk assessment of chemicals. Look-up tables and flowcharts were developed. A second example discusses the use of DEB-IBM as refinement tool for laboratory-to-field extrapolations. The effect of food density in combination with lethal and and sub-lethal effects to D. magna populations was investigated. As final example, DEB-IBM was linked to FOCUS (a dedicated exposure model that predicts the fate of pesticides in the environment) to predicted realistic effects of pesticide mixtures to D. magna populations. A realistic example was developed with a water body contaminated with endosulfan and funguren (a copper pesticide) due to pesticide application on nearby fields. The predicted surface water concentrations from FOCUS were linked with DEB-IBM. In addition, the DEB-IBM predictions were compared to a traditional dose-response curve analysis and predictions with a TKTD model. Good model documentation and accessibility is required to increase model transparency and reliability. An extensive description of the model, following the TRACE (transparent and comprehensive model ‘evaludation’) documentation, is provided (Appendix E). We conclude that mechanistic population models can be used for prospective and predictive risk assessment of chemical mixtures. More so than predicting effects, mechanistic population models can also give information and understanding of the driving forces of mixture toxicity within a population context. However, there is still a lack of guidance on the ‘standardized’ use of these models. More applications and communication of results would help increase acceptance of mechanistic population models for regulatory risk assessment. With this thesis we have shown that mechanistic population models can bridge multiple uncertainty gaps that were previously unaddressed in ecological risk assessment: the divide between individuals and populations, between single substances and mixtures of substances, and between constant controlled exposure conditions and dynamic exposure conditions.