Background: While epithelial–mesenchymal transition (EMT) can be readily induced experimentally in cancer cells, the EMT process as manifested in human tumors needs to be better understood. Pan‐cancer genomic datasets from The Cancer Genome Atlas (TCGA), representing over 10,000 patients and 32 distinct cancer types, provide a rich resource for examining correlative patterns involving EMT mediators in the setting of human cancers. Results: Here, we surveyed a 16‐gene signature of canonical EMT markers in TCGA pan‐cancer cohort. The histology or cell‐of‐origin of a tumor sample may align more with mesenchymal or epithelial phenotype, and noncancer as well as cancer cells can contribute to the overall molecular patterns observed within a tumor sample; correlation models involving EMT markers can factor in both of the above variables. EMT‐associated genes appear coordinately expressed across all cancers and within each cancer type surveyed. Gene signatures of immune cells correlate highly with EMT marker expression in tumors. In pan‐cancer analysis, several EMT‐related genes can be significantly associated with worse patient outcome. Conclusions: Gene correlates of EMT phenotype in human tumors could include novel mediators of EMT that might be confirmed experimentally, by which TCGA datasets may serve as a platform for discovery in ongoing studies. Developmental Dynamics 247:555‐564, 2018. © 2017 Wiley Periodicals, Inc. Key Findings Survey of expression patterns for canonical EMT markers in TCGA pan-cancer cohort, representing over 10,000 patients and 32 distinct cancer types. For multiple cancer types, gene signatures of immune cells are found to correlate highly with EMT marker expression in human tumors. Several EMT‐related genes can be significantly associated with worse patient outcome (time to death), independent of cancer type.