This study introduces LEfSe, a method for metagenomic biomarker discovery and explanation. LEfSe combines statistical significance tests with biological consistency and effect size estimation to identify features (organisms, genes, or pathways) that consistently explain differences between microbial communities. The method is validated on multiple microbiomes and provides an online interface at http://huttenhower.sph.harvard.edu/lefse/. Biomarker discovery is crucial for translating molecular data into clinical applications. Metagenomic studies have shown that microbial communities can serve as biomarkers for host factors such as lifestyle and disease. However, identifying consistent biomarkers remains challenging due to high-dimensional data and the need for biological interpretation. LEfSe addresses these challenges by using linear discriminant analysis (LDA) to estimate effect sizes and identify features that are both statistically significant and biologically relevant. It first identifies features that are differentially abundant between classes and then tests for biological consistency. LEfSe also provides a visualization of biomarkers on taxonomic trees, which helps in summarizing results in a biologically meaningful way. The method was validated using human microbiome data, a mouse model of ulcerative colitis, and environmental samples. LEfSe was also tested on synthetic data, showing a lower false positive rate compared to standard statistical tests, at the cost of a slightly higher false negative rate. Results showed that LEfSe effectively identified differentially abundant features in the human microbiome, including mucosal and aerobic taxa. It also detected specific microbial clades in different environments, such as the gut, oral cavity, and skin. LEfSe was able to distinguish between different microbial communities, including those in a mouse model of colitis, and identified specific biomarkers associated with disease states. In addition, LEfSe was compared with other metagenomic analysis tools, such as Metastats and the KW test. LEfSe demonstrated a lower false positive rate and better performance in detecting biomarkers with consistent biological explanations. It also provided a more comprehensive analysis of functional roles in microbial communities, including the identification of pathways and biological mechanisms over- or under-represented in different communities. The study highlights the importance of integrating statistical and biological significance in metagenomic biomarker discovery. LEfSe provides a robust framework for identifying and explaining biomarkers in microbial communities, which can be applied to various fields, including microbiology, ecology, and clinical applications. The method is freely available online and can be used for both real and synthetic data analysis.This study introduces LEfSe, a method for metagenomic biomarker discovery and explanation. LEfSe combines statistical significance tests with biological consistency and effect size estimation to identify features (organisms, genes, or pathways) that consistently explain differences between microbial communities. The method is validated on multiple microbiomes and provides an online interface at http://huttenhower.sph.harvard.edu/lefse/. Biomarker discovery is crucial for translating molecular data into clinical applications. Metagenomic studies have shown that microbial communities can serve as biomarkers for host factors such as lifestyle and disease. However, identifying consistent biomarkers remains challenging due to high-dimensional data and the need for biological interpretation. LEfSe addresses these challenges by using linear discriminant analysis (LDA) to estimate effect sizes and identify features that are both statistically significant and biologically relevant. It first identifies features that are differentially abundant between classes and then tests for biological consistency. LEfSe also provides a visualization of biomarkers on taxonomic trees, which helps in summarizing results in a biologically meaningful way. The method was validated using human microbiome data, a mouse model of ulcerative colitis, and environmental samples. LEfSe was also tested on synthetic data, showing a lower false positive rate compared to standard statistical tests, at the cost of a slightly higher false negative rate. Results showed that LEfSe effectively identified differentially abundant features in the human microbiome, including mucosal and aerobic taxa. It also detected specific microbial clades in different environments, such as the gut, oral cavity, and skin. LEfSe was able to distinguish between different microbial communities, including those in a mouse model of colitis, and identified specific biomarkers associated with disease states. In addition, LEfSe was compared with other metagenomic analysis tools, such as Metastats and the KW test. LEfSe demonstrated a lower false positive rate and better performance in detecting biomarkers with consistent biological explanations. It also provided a more comprehensive analysis of functional roles in microbial communities, including the identification of pathways and biological mechanisms over- or under-represented in different communities. The study highlights the importance of integrating statistical and biological significance in metagenomic biomarker discovery. LEfSe provides a robust framework for identifying and explaining biomarkers in microbial communities, which can be applied to various fields, including microbiology, ecology, and clinical applications. The method is freely available online and can be used for both real and synthetic data analysis.