A lack of clinically relevant experimental models of human prostate cancer hampers evaluation of potential therapeutic agents. Currently, androgen deprivation therapy is the gold standard treatment for advanced prostate cancer, but inevitably a sub-population of cancer cells survive and repopulate the tumor. Tumor cells that survive androgen withdrawal are critical therapeutic targets for more effective treatments but current model systems cannot determine when they arise in disease progression and are unable to recapitulate variable patient response to treatment. In this study a model system was developed in which stromal-supported xenografts from 12 patients with early-stage localized disease can be tested for response to castration. The histopathology of these xenografts mimicked the original tumors, and short-term host castration resulted in reduced proliferation and increased apoptosis in tumor cells. After 4 weeks of castration, residual populations of quiescent, stem-like tumor cells remained. Without subsequent treatment, these residual cells displayed regenerative potential, as testosterone re-administration resulted in emergence of rapidly proliferating tumors. AR- and AR+ stromal lines (PShTert and PShTertAR) were used to demonstrate that low stromal AR augmented the response to 3 days of castration resulting in a decreased apoptotic response in xenografted tumors. In summary, this model may be useful for revealing potential cellular targets in prostate cancer, which exist prior to the onset of aggressive incurable disease, and mechanisms with which to target them. Specific eradication of these regenerative stem-like tumor cells that evade castration therapy could improve patient survival.