Epithelial-mesenchymal transition is characterized by a loss of cell adhesion and increased cell mobility due to cells gaining a mesenchymal phenotype. During the epithelial-mesenchymal transition process, tumor cells are expected to lose their epithelial phenotype and gradually and sequentially acquire a mesenchymal phenotype. Epithelial-mesenchymal transition is a dynamic and reversible process, which has been observed in patient tissues to display a wide spectrum of phenotypes. However, very little is known about the clinical significance of the different phenotypes of the epithelial-mesenchymal transition. Based on the expression pattern of various epithelial-mesenchymal transition-related proteins, we divided 168 esophageal squamous cell carcinomas into different phenotypes, including complete type; incomplete type, including hybrid type and null type; and a wild type. The clinical significance of each phenotype was investigated. Of the 168 cases, 31 were categorized as complete type, 53 as incomplete type (hybrid type, 26 cases; null type, 27 cases), and 84 as wild type. Epithelial-mesenchymal transition phenotype was significantly associated with tumor size (P ¼ 0.021), differentiation (P ¼ 0.001), and invasion depth (Po0.001). Overall survival and disease-free survival rates were significantly worse in the complete type, better in the incomplete type, and best in the wild type. Within the incomplete type group, the hybrid type survival curve was similar to that of the complete type, whereas the overall survival of the null type was similar to the wild type. Complete type had a noticeable poorer prognostic effect on survival in patients with early invasion (pTr2) than it had on survival among patients with advanced invasion (pTZ3). The complete phenotype was an independent prognostic factor for both overall (P ¼ 0.009) and disease-free survival (Po0.001). In conclusion, classification of epithelial-mesenchymal transition phenotypes has novel clinical implications, and identification of a specific phenotype might provide a tool to better stratify and predict patient outcomes.