The estimation of ecotoxicological properties of chemicals is a major environmental concern. The QSAR models (Quantitative Structure-Activity Relationship) are linear regression and classification models often used to predict the ecotoxicity of che- mical molecules. We consider in this paper quantile regression estimators which are more robust to outliers providing a more detailed focus on the entire conditional distribution of the dependent variable and not only on its mean as in linear regression. We propose here, in this concern of prediction, quantile models in regression and Support Vector Machines (SVM) in the field of chemoinformatics.