Epithelial-mesenchymal transition (EMT) has been implicated as the critical event initiating cancer invasion and metastasis. After disseminating through the circulation, the malignant cells have been proposed to undergo subsequent mesenchymal-epithelial transition (MET) to form secondary tumors. However, strong evidence from human tumor specimens for this paradigm is lacking. In carcinomas, cancers derived from epithelial tissues, epithelial morphology and gene expression are always retained to some degree. While mesenchymal transdifferentiation may be involved in the pathogenesis of carcinosarcomas, even in these neoplasms, as well as in germ cell tumors capable of multilineage differentiation, the mesenchymal phenotype does not facilitate metastatic progression. Indeed, most cancers invade and travel through lymphatic and blood vessels via cohesive epithelial migration, rather than going through the EMT-MET sequence. EMT gene expression is also consistently associated with high histologic grade and while the transcription factors, Snail, Slug and Twist have traditionally been thought of as inducers of EMT, under certain conditions, they also mediate dedifferentiation and maintenance of the stem cell state. In various malignancies, including basal-like breast cancer and colorectal cancer, the genetically unstable, undifferentiated phenotype predicts early metastatic spread and poor prognosis. This article discusses some of the controversies surrounding differentiation and metastasis from a clinicopathologic perspective and presents evidence that the epithelial phenotype is maintained throughout the process of cancer metastasis.