Medulloblastoma (MB) is the most common malignant pediatric brain tumour, and is categorized into four molecular subgroups, with Group 3 MB having the worst prognosis due to the highest rate of metastatic dissemination and relapse. In this work, we describe the epigenetic regulator Bmi1 as a novel therapeutic target for treatment of recurrent Group 3 MB. Through comparative profiling of primary and recurrent MB, we show that Bmi1 defines a treatment-refractory cell population that is uniquely targetable by a novel class of small molecule inhibitors. We have optimized an in vivo mouse-adapted therapy model that has the advantage of generating recurrent, human, treatment-refractory MBs. Our preliminary studies showed that although chemoradiotherapy administered to mice engrafted with human MB showed reduction in tumour size, Bmi1 expression was enriched in the post-treatment residual tumour. Furthermore, we found that knockdown of Bmi1 in human recurrent MB cells decreases proliferation and self-renewing capacities of MB cells in vitro as well as both tumour size and extent of spinal leptomeningeal metastases in vivo. Oral administration of a potent Bmi1 inhibitor, PTC 028, resulted in a marked reduction in tumour burden and an increased survival in treatment cohort. Bmi1 inhibitors showed high specificity for MB cells and spared normal human neural stem cells, when treated with doses relevant for MB cells. As Group 3 medulloblastoma is often metastatic and uniformly fatal at recurrence, with no current or planned trials of targeted therapy, an efficacious agent such as Bmi1 inhibitor could be rapidly transitioned to clinical trials.