The paper introduces MuTect, a method for detecting somatic point mutations in cancer samples, particularly those that are impure or heterogeneous. MuTect applies a Bayesian classifier to identify mutations with very low allele fractions, followed by filters to ensure high specificity. The authors describe benchmarking approaches using real sequencing data to evaluate sensitivity and specificity as functions of sequencing depth, base quality, and allelic fraction. Compared to other methods, MuTect shows higher sensitivity, especially for mutations with allelic fractions as low as 0.1, making it useful for studying cancer subclones and their evolution in standard exome and genome sequencing data. The method's performance is supported by independent experimental validation and its application to various datasets. MuTect is freely available for noncommercial use.The paper introduces MuTect, a method for detecting somatic point mutations in cancer samples, particularly those that are impure or heterogeneous. MuTect applies a Bayesian classifier to identify mutations with very low allele fractions, followed by filters to ensure high specificity. The authors describe benchmarking approaches using real sequencing data to evaluate sensitivity and specificity as functions of sequencing depth, base quality, and allelic fraction. Compared to other methods, MuTect shows higher sensitivity, especially for mutations with allelic fractions as low as 0.1, making it useful for studying cancer subclones and their evolution in standard exome and genome sequencing data. The method's performance is supported by independent experimental validation and its application to various datasets. MuTect is freely available for noncommercial use.