Machine learning has become a central feature in the development or refinement of in silico methodologies and techniques. Quantitative structure-activity relationship (QSAR) models are no exception. In fact, one can consider there is a renaissance of QSAR techniques and respective reliability as there is a greater synergy between the two of them. Further, this new wave of QSAR + machine learning (ML) techniques allows new avenues in several fields of application, namely, when regarding cytotoxicity and/or ecotoxicity monitoring of nanoparticles (NPs). The latter is of major importance, as the challenges brought by environment management and the increasing concern it has on the food chain are met with expensive and overall slow experimental answers. Within this context, and alongside classical QSAR + machine learning techniques, recent QSAR perturbation-based models join methods with ML as well. The QSAR perturbation models feature the possibility of simultaneous modeling multi bio-targets versus NPs in different experimental conditions, thus offering practical solutions to classical QSAR + ML limitations. The use of in silico models could be the most feasible answer to the present and future scenarios of mandatory ecotoxicity monitorization for nanotechnology by-products. This chapter approaches the methodologies and fundamentals of classical and perturbation-based QSAR models within the environmental risk assessment framework, as scaffold to develop novel in silico techniques.