The selection of compounds with a similar toxicological mode of action is a key problem in the study of chemical mixtures. In this paper, an approach for the selection of chemicals with similar mode of action, based on the analysis of structural similarities by means of QSAR and chemometric methods, is described. As a first step, a complete representation of chemical structures for examined chemicals (phenylureas and triazines) by different sets of molecular descriptors allows a preliminary exploration of similarity using multi-dimensional scaling (MDS). The use of genetic algorithm (GA) to select the most relevant molecular descriptors in modeling toxicity data makes it possible to develop predictive toxicity models. The final step is a similarity analysis, based again on MDS, using selected molecular descriptors, really relevant in describing the toxicological effect.