Abstract: Qualitative classification and quantitative regression models for fathead minnow and honey bee toxicity prediction were developed using different chemoinformatics techniques such as substructure pattern recognition and different machine learning methods.Specifically,methods include support vector machine,C4.5 decision tree,k-nearest neighbors,random forest and naive bayes.Reliable predictive models were developed and all models were validated by the independent test set.The overall predictive accuracy of the classification models using support vector machine were 95.9% for the fathead minnow test set and 95.0% for the honey bee test set.The square of correlation coefficient of regression models were 0.878 for the fathead minnow test set and 0.663 for the honey bee test set using support vector machine regression algorithm.At last,some representative substructure patterns for characterizing fathead minnow and honey bee toxicity compounds,such as 1,2-diphenol,dialkylthioether,diarylether and phosphoric_ acid_ derivative were also identified via the information gain analysis.The approaches provide a useful strategy and robust tools in the screening of ecotoxicological risk or environmental hazard potential of chemicals.