Unavailability of clean and fresh water will surely become one of the greatest mankind problems if steady development of global economy and the population growth are going to continue. Various odd substances have been found in waters in the past few decades: they were found in wastewater, surface water or even drinking water; the devastating fact is that the origin of these substances is mostly related to human activities. These water pollutants include various industrial chemicals, pharmaceuticals, pesticides, metals, etc. Among them, pharmaceuticals and pesticides have been identified recently as compounds with severe hazardous potential for humans, flora and fauna, since these substances are aimed to interact with living beings. Accordingly, the related risk assessment is required by regulatory bodies. The risk assessment is mostly based on information about toxicities of single pollutants. Unfortunately, such approach has a considerable defect. Namely, situation when only one pollutant is present in the environment is generally very rare; the living beings are mostly exposed to mixtures of pollutants. In such mixtures, a joined activity of substances is possible, such which might have more hazardous influence then the both substances acting alone. Therefore, risk assessment based on information about mixture toxicity is much more acceptable approach. The number of possible mixtures is infinite and, accordingly, it is impossible to experimentally determine toxicity for each mixture. Evermore, the number of newly-created substances in the environment is also increasing and for such substances there is mostly no toxicity information. The scientists are trying to solve this problem by chemometrics or more precisely by the application of various toxicity models. Two most common models are: Concentration Addition model (CA) and Independent Action model (IA). The rapid development of computer technology allowed for the application of various advanced computational methods and tools. The one with great potential is the so-called QSAR approach (Quantitative Structure-Activity Relationship) which correlates the structural characteristics of the substances with their certain activity, for example toxicity. The aim of this dissertation was: I. to classify the modes of joint toxic activity of selected pharmaceuticals and pesticides, and II. to develop QSAR models which would provide prediction of mixture toxicity based on information about mixture composition and of molecular structures of its constituents. Binary mixtures of substances contained in molar ratios of 25:75, 50:50 and 75:25 were tested in this work. Toxicities were determined according to ISO 11348 by measuring the luminescence of Vibrio fischeri bacteria; they were expressed on three levels: as EC50, EC30 and EC10 values. Ten pollutants were selected for the analysis of joint toxic activity; this referred to six pharmaceuticals: azithromycin, erythromycin, carbamazepine, oxytetracycline, dexamethasone and diclofenac, and four pesticides: alachlor, isoproturon, diuron and chlorfenvinphos. Toxic activities were studied on binary mixtures: mixtures of pharmaceuticals and mixtures of pesticides were analyzed separately. CA and IA toxicity models were applied for the analysis of joint toxic activity and the model-predicted values were compared to experimentally determined ones. Toxicities of single-pollutant solutions were used as inputs to the models and joint toxic activities were classified as: I. the additive behavior, and II. synergistic or antagonistic deviation from the additive behavior. Deviations from the additive action were confirmed additionally through better fitting of the experimental values in the case of IA model. The results indicated synergistic behavior in the mixture of diclofenac and oxytetracycline. Other binary combinations of diclofenac and oxytetracycline were acting similarly: thus the synergism was observed for their mixtures with carbamazepine, while the antagonistic behavior was shown for the mixtures with azithromycin and erythromycin. All the remaining binary mixtures of pharmaceuticals indicated additive behavior. The analysis revealed additive behavior for all binary mixtures of pesticides; the exception was the mixture of isoproturon and chlorfenvinphos whose combination resulted with synergistic deviation from the additive behavior. In the last stage of the research, QSAR models were developed and validated. Four new pharmaceuticals were added to the study: omeprazole, desloratadine, imatinib, and tobramycin, as well as their binary mixtures. The intention was to increase the number of data for development and validation of QSAR models. Molecular conformations of all selected pollutants were optimized and the descriptors were calculated. Different mixing rules were tested in order to find the best way to present mathematically the binary mixtures. These rules included various ways of combining molecular descriptor values with mixture composition. Modeling was performed by Multiple Linear Regression and Genetic Algorithm was applied in the selection of most-informative descriptors. QSAR models were developed for each of the three tested toxicity levels: EC50, EC30 and EC10. The models were validated using the external data set, i.e. by the mixtures whose toxicities were not included in development of the models. Each of the models showed a high level of accuracy. QSAR models were used in revealing the structural characteristics that had a significant impact on toxicity towards Vibrio fischeri. The results indicated that EC50 and EC30 values were mostly influenced by geometrical distances between nitrogen and sulfur atoms. Additionally, the simultaneous presence of oxygen and chlorine atoms could induce the increase in mixture toxicity. At the lowest tested level (EC10), dipole moment had the highest impact on the mixture toxicity.