The primary cause of cancer-related deaths is metastasis— the spread of cancer cells to distant organs— and yet the mechanisms underlying this process remain elusive due to the difficulty in detecting early metastatic events, which are rare, stochastic and microscopic. To investigate the cellular and molecular mechanisms of metastasis, I utilized an autochthonous mouse model of pancreatic cancer (KPCY) in which all tumor cells are genetically labeled with yellow fluorescent protein (YFP). The YFP lineage label allows for the detection and isolation of disseminated tumor cells as they delaminate from epithelial structures within the primary tumor, invade into the stroma and circulation, and colonize distal organs. Using this system, I characterized the development of metastatic lesions from single disseminated cells to grossly macroscopic lesions in the murine liver. I found that gross metastases closely resembled primary tumors in terms of differentiation and microenvironment— these large lesions are well differentiated, containing primarily epithelial tumor cells, and accumulate stroma consisting of myofibroblasts, leukocytes and extracellular matrix (ECM). In contrast, single disseminated cells tend to be poorly differentiated and lack any association with stromal cells, and must build up a microenvironment around them as they grow. Despite the presumably protective stroma surrounding large lesions, gross metastasis was significantly reduced with chemotherapy, while single cells were unaffected. Interestingly, residual lesions were enriched for epithelial features, suggesting that EMT confers chemosensitivity in this context. I also used the KPCY model to investigate the molecular mechanisms of epithelial-mesenchymal transition (EMT), which is widely considered to be the first step in the metastatic cascade. The YFP lineage label made it possible to identify and isolate tumor cells that have undergone EMT for transcriptional profiling. Surprisingly, I found that in a majority of pancreatic tumors, conventional transcriptional repressors were not involved in EMT. Although a mesenchymal transcriptional program was significantly enriched in cells that had undergone this “non-canonical” mechanism of EMT, the epithelial program was downregulated at the protein level by a mechanism involving protein internalization. Because cells retain both epithelial and mesenchymal properties during non-canonical EMT, this phenomenon represents an attractive explanation for the ability of tumor cells to cycle between epithelial and mesenchymal states and adapt to the changing microenvironment on their way to metastatic sites. The journey from primary tumor to metastatic site requires cancer cells to overcome many obstacles and a better understanding of how they navigate the numerous steps of the metastatic cascade could open the door to desperately needed anti-metastatic therapies.