In our studies of ovarian cancer cells we have identified subpopulations of cells that are in a transitory E/M hybrid stage, i.e. cells that simultaneously express epithelial and mesenchymal markers. E/M cells are not homogenous but, in vitro and in vivo, contain subsets that can be distinguished based on a number of phenotypic features, including the subcellular localization of E-cadherin, and the expression levels of Tie2, CD133, and CD44. A cellular subset (E/M-MP) (membrane E-cadherin(low)/cytoplasmic E-cadherin(high)/CD133(high), CD44(high), Tie2(low)) is highly enriched for tumor-forming cells and displays features which are generally associated with cancer stem cells. Our data suggest that E/M-MP cells are able to differentiate into different lineages under certain conditions, and have the capacity for self-renewal, i.e. to maintain a subset of undifferentiated E/M-MP cells during differentiation. Trans-differentiation of E/M-MP cells into mesenchymal or epithelial cells is associated with a loss of stem cell markers and tumorigenicity. In vivo xenograft tumor growth is driven by E/M-MP cells, which give rise to epithelial ovarian cancer cells. In contrast, in vitro, we found that E/M-MP cells differentiate into mesenchymal cells, in a process that involves pathways associated with an epithelial-to-mesenchymal transition. We also detected phenotypic plasticity that was dependent on external factors such as stress created by starvation or contact with either epithelial or mesenchymal cells in co-cultures. Our study provides a better understanding of the phenotypic complexity of ovarian cancer and has implications for ovarian cancer therapy.