The authors present a new method and software tool, PolyPhen-2, for predicting the impact of protein sequence variants. PolyPhen-2 uses a combination of sequence-based and structure-based predictive features, selected by an iterative greedy algorithm. The tool employs a Naïve Bayes classifier to predict the functional significance of amino acid replacements. Two datasets, HumDiv and HumVar3, were used to train and test PolyPhen-2, with the former focusing on damaging mutations causing human Mendelian diseases and the latter on all human disease-causing mutations. PolyPhen-2 outperforms its predecessor, PolyPhen1, and other popular prediction tools in terms of accuracy. The tool provides posterior probabilities and qualitative assessments of mutations, making it suitable for different types of genetic studies, such as diagnostics of Mendelian diseases and analysis of complex phenotypes.The authors present a new method and software tool, PolyPhen-2, for predicting the impact of protein sequence variants. PolyPhen-2 uses a combination of sequence-based and structure-based predictive features, selected by an iterative greedy algorithm. The tool employs a Naïve Bayes classifier to predict the functional significance of amino acid replacements. Two datasets, HumDiv and HumVar3, were used to train and test PolyPhen-2, with the former focusing on damaging mutations causing human Mendelian diseases and the latter on all human disease-causing mutations. PolyPhen-2 outperforms its predecessor, PolyPhen1, and other popular prediction tools in terms of accuracy. The tool provides posterior probabilities and qualitative assessments of mutations, making it suitable for different types of genetic studies, such as diagnostics of Mendelian diseases and analysis of complex phenotypes.