Metastasis is the most lethal and insidious aspect of cancer. Despite significant therapeutic advances, metastatic disease is generally incurable. To date, the natural history, clonal evolution and patterns of systemic spread are poorly understood, hindering effective treatment and prevention efforts. In particular, it remains difficult to predict which patients will relapse at the time of diagnosis. In this talk, I will describe several quantitative frameworks to delineate the dynamics of human tumor progression and metastasis and their application to diverse cancer genome sequencing data. I will begin by outlining a suite of computational tools to infer the evolutionary dynamics of tumor progression from patient genomic data by coupling population genetic theory, spatial computational modeling and approximate Bayesian computation. In particular, I will describe a new method to infer the timing of metastasis based on patterns of genomic divergence between paired primary tumors and distant metastases and show how application of this approach yields the first quantitative evidence for early systemic spread in colorectal cancer and insights into the genomic drivers of this lethal process. Building on these findings, I will describe the analysis of a large series of breast, colon and lung cancer patients, providing further evidence for early metastatic spread in these common cancer types, while revealing patterns of metastatic seeding and the profound impact of therapy. Lastly, I will describe a statistical approach to model the dynamics of breast cancer relapse and its application to a cohort of nearly 2000 breast cancers with detailed long term clinical follow-up and genomic information. Throughout, I will discuss the context dependencies that underlie disease progression and how this may inform strategies for patient stratification and therapeutic targeting.