Brain Metastases (BM) represent a leading cause of cancer mortality. While metastatic lesions contain subclones derived from their primary lesion, their functional characterization has been limited by a paucity of preclinical models accurately recapitulating the stages of metastasis. This work describes the isolation of a unique subset of metastatic stem-like cells from primary human patient samples of BM, termed brain metastasis initiating cells (BMICs). Utilizing these BMICs we have established a novel patient-derived xenograft (PDX) model of BM that recapitulates the entire metastatic cascade, from primary tumor initiation to micro-metastasis and macro-metastasis formation in the brain. We then comprehensively interrogated human BM to identify genetic regulators of BMICs using in vitro and in vivo RNA interference screens, and validated hits using both our novel PDX model as well as primary clinical BM specimens. We identified SPOCK1 and TWIST2 as novel BMIC regulators, where in our model SPOCK1 regulated BMIC self-renewal and tumor initiation, and TWIST2 specifically regulated cell migration from lung to brain. A prospective cohort of primary lung cancer specimens was used to establish that SPOCK1 and TWIST2 were only expressed in patients who ultimately developed BM, thus establishing both clinical and functional utility for these gene products. This work offers the first comprehensive preclinical model of human brain metastasis for further characterization of therapeutic targets, identification of predictive biomarkers, and subsequent prophylactic treatment of patients most likely to develop BM. By blocking this process, metastatic lung cancer would effectively become a localized, more manageable disease.