Dataset D1 comprises information extracted from anonymized hematoxylin and eosin (H & E) stained needle-core biopsies of prostate tissue digitized at 20x optical magnification on a whole-slide digital scanner. Regions corresponding to prostate cancer and normal tissue were manually delineated by a pathologist on digitized images. Each image was divided into non-overlapping 30 × 30-pixel tissue regions ("image patches"). An equal number of image patches were then sampled from cancerous (label=1) and normal (label=-1) annotated regions. The subset of data used for error rate extrapolation comprises 100 image patches. The larger dataset used for error rate validation comprises 500 image patches. Feature information for these data comprises the 2 most important descriptors (selected via mRMR) from a larger set of statistical, co-occurrence, and filter responses extracted from each image patch.