CIBERSORT is a computational method for characterizing cell composition in complex tissues based on gene expression profiles. It outperforms other methods in handling noise, unknown mixture content, and closely related cell types. CIBERSORT uses a novel application of linear support vector regression (SVR) to estimate relative proportions of cell types in gene expression profiles. It requires a reference gene expression signature matrix, such as LM22, which contains 547 genes distinguishing 22 human hematopoietic cell phenotypes. CIBERSORT was validated on various datasets, including external leukocyte subsets and human tonsils, showing high accuracy in identifying cell types. It also demonstrated strong performance in distinguishing positive from negative samples with high sensitivity and specificity. CIBERSORT was benchmarked against six GEP deconvolution methods, showing superior accuracy in mixtures with unknown content and noise. It was applied to formalin-fixed, paraffin-embedded (FFPE) samples and showed significant correlation with flow cytometry measurements. CIBERSORT was also tested on microarray and RNA-Seq data, demonstrating its utility for analyzing cellular heterogeneity in various tissue types. The method is robust to noise and multicollinearity, and does not require cell type-specific expression for every gene, allowing for broader applicability. CIBERSORT is available online at http://cibersort.stanford.edu. The method has limitations, including the fidelity of reference profiles and the need for further refinement in detection limits for individual cell types. Overall, CIBERSORT provides a powerful tool for analyzing cell composition in complex tissues.CIBERSORT is a computational method for characterizing cell composition in complex tissues based on gene expression profiles. It outperforms other methods in handling noise, unknown mixture content, and closely related cell types. CIBERSORT uses a novel application of linear support vector regression (SVR) to estimate relative proportions of cell types in gene expression profiles. It requires a reference gene expression signature matrix, such as LM22, which contains 547 genes distinguishing 22 human hematopoietic cell phenotypes. CIBERSORT was validated on various datasets, including external leukocyte subsets and human tonsils, showing high accuracy in identifying cell types. It also demonstrated strong performance in distinguishing positive from negative samples with high sensitivity and specificity. CIBERSORT was benchmarked against six GEP deconvolution methods, showing superior accuracy in mixtures with unknown content and noise. It was applied to formalin-fixed, paraffin-embedded (FFPE) samples and showed significant correlation with flow cytometry measurements. CIBERSORT was also tested on microarray and RNA-Seq data, demonstrating its utility for analyzing cellular heterogeneity in various tissue types. The method is robust to noise and multicollinearity, and does not require cell type-specific expression for every gene, allowing for broader applicability. CIBERSORT is available online at http://cibersort.stanford.edu. The method has limitations, including the fidelity of reference profiles and the need for further refinement in detection limits for individual cell types. Overall, CIBERSORT provides a powerful tool for analyzing cell composition in complex tissues.