Environmental Risk assessment (ERA) of the more than 100,000 chemical compounds on the European Inventory of Existing Chemicals (EINECS List) is a tasks beyond human, technical and economical resources. An important tool to overcome this hurdle is to use Quantitative StructureActivity Relationships (QSAR). QSARs are models that quantify endpoints, e.g. physico-chemical properties as well as fixed toxicity parameters, as a function of inherent molecular property descriptors. As such, these are multi-pollutant models that may be used for supplying data of identified hazardous pollutants, where experimental measurements are missing. In this way QSARs may be the link to overcome the backlog with respects to the number of chemicals, which are to be assessed by the EU member states. In this dissertation, the limitations in the use of QSARs for predicting endpoints, such as partitioning coefficients between different natural occurring phases and fixed toxicity endpoints, for use in ERA is investigated. Furthermore, the potential of multivariate SAR and QSAR for increasing the knowledge of significant structural and electronic intrinsic molecular properties explaining the variations in endpoint values is investigated. To overcome the limitations in the application of QSARs for supplying data for ERA, as well as for gaining knowledge concerning mechanisms and significant parameters determining the potential hazards of environmental pollutants, an elucidation of uncertainties and unknown parameters which affect the measured endpoint is needed. Simple endpoints such as the aqueous solubility and octanol-water partition coefficients show significant variations between experimental standard methods and specific experimental conditions in the measured system. In this respects, quality assurance, or preprocessing, of the data used for calibrating QSARs, as well as process understanding with respect to the measured system, are shown be crucial for the predictive power of QSARs. In this dissertation different aspects with respect to the development of scientific valid QSAR are identified: 1) The informational content of the empirical versus the non-empirical and quantumchemical descriptors has been evaluated, 2) The performance of QSARs based on simple linear regression (LR) and partial least square regression (PLS) have been investigated, and 3) The importance of the quality of data, as well as understanding of experimental/environmental measured systems, to be modelled have been elucidated. The present ERA concept, as well as the paradigm of QSARs, are based on substance specific properties only and do not include any effects from variations in the nature and characteristics of the natural phases, e.g. soil, sediment and water, in which the pollutants occur. This dissertation focuses on inconsistency and uncertainties in measured endpoints, which result in additional unknown parameters included in the calibration of QSARs. Through investigations of the additional nonquantified uncertainties, or known but not included background data, the quality of data used in the development of QSARs is shown to be critical for the robustness and validity of QSARs. Main aspects are shown to explain the variability in endpoint data found in the literature. These are 1) significant influences of background data, i.e., environmental or experimental parameters such as pH and temperature, 2) the presence of dissolved organic matter (DOM), and 3) the thermodynamic equilibrium description of the pollutant and phases of one and multi-phase systems, when quantifying physico-chemical properties of organic hydrophobic substances. ii Through the use of classical statistics as well as multivariate data analysis, the quality of data, interpretations of informational content and model performances of QSARs have been evaluated. Furthermore, the influence of environmental parameters, e.g. pH, temperature, solutions vs. mixtures, and dissolved organic matter (DOM), on the model performance of QSARs have been analysed. The most critical aspects with respect to the development of scientific valid QSARs seems not to be the model concepts, but high uncertainties and inconsistency in the data used for calibrating QSARs. Concepts of how to overcome the critical aspects and thus make substantial improvements in the applicability of QSARs are proposed.