Motivation: Expression profile of normal tissue is primary source to find genes showing aberrant expression pattern specific in matched cancer tissue, but sample number of normal control in public gene expression repositories is disproportionally small compared to cancer and scattered in several datasets. Results: We built oncopression by integrating several datasets into one large dataset for comprehensive analysis about 25 types of human cancers including 20 640 cancer samples and 6801 normal control profiles. Expression profiles in cancers can be directly compared to normal tissue counterparts. Validity of the integration was tested using immunohistochemical staining results and principal component analysis. We have utilized the pre-release version of oncopression to identify cancer-specific genes in several studies. Availability and Implementation: Free access at http://www. oncopression. com and all expression data are available for download at the site.