Over 90% of all human neoplasia is derived from epithelia. Significant progress has been made in the identification of stem cells of many epithelia. In general, epithelial stem cells lack differentiation markers, have superior in vivo and in vitro proliferative potential, form clusters in association with a specialized mesenchymal environment (the 'niche'), are located in well-protected and nourished sites, and are slow-cycling and thus can be experimentally identified as 'label-retaining cells'. Stem cells may divide symmetrically giving rise to two identical stem cell progeny. Any stem cells in the niche, which defines the size of the stem cell pool, may be randomly expelled from the niche due to population pressure (the stochastic model). Alternatively, a stem cell may divide asymmetrically yielding one stem cell and one non-stem cell that is destined to exit from the stem cell niche (asymmetric division model). Stem cells separated from their niche lose their stemness, although such a loss may be reversible, becoming 'transit-amplifying cells' that are rapidly proliferating but have a more limited proliferative potential, and can give rise to terminally differentiated cells. The identification of the stem cell subpopulation in a normal epithelium leads to a better understanding of many previously enigmatic properties of an epithelium including the preferential sites of carcinoma formation, as exemplified by the almost exclusive association of corneal epithelial carcinoma with the limbus, the corneal epithelial stem cell zone. Being long-term residents in an epithelium, stem cells are uniquely susceptible to the accumulation of multiple, oncogenic changes giving rise to tumors. The application of the stem cell concept can explain many important carcinoma features including the clonal origin and heterogeneity of tumors, the occasional formation of tumors from the transit amplifying cells or progenitor cells, the formation of precancerous 'patches' and 'fields', the mesenchymal influence on carcinoma formation and behavior, and the plasticity of tumor cells. While the concept of cancer stem cells is extremely useful and it is generally assumed that such cells are derived from normal stem cells, more work is needed to identify and characterize epithelial cancer stem cells, to address their precise relationship with normal stem cells, to study their markers and their proliferative and differentiation properties and to design new therapies that can overcome their unusual resistance to chemotherapy and other conventional tumor modalities.