Graft-versus-host disease (GVHD) is a life-threatening complication of allogeneic hematopoietic cell transplantations (allo-HCT) used for the treatment of hematological malignancies and other blood-related disorders. Until recently, the discovery of actionable molecular targets to treat GVHD and their preclinical testing was almost exclusively based on modeling allo-HCT in mice by transplanting bone marrow and splenocytes from donor mice into MHC-mismatched recipient animals. However, due to fundamental differences between human and mouse immunology, the translation of these molecular targets into the clinic can be limited. Therefore, humanized mouse models of GVHD were developed to circumvent this limitation. In these models, following the transplantation of human peripheral blood mononuclear cells (PBMCs) into immunodeficient mice, T cells recognize and attack mouse organs, inducing GVHD. Thereby, humanized mice provide a platform for the evaluation of the effects of candidate therapies on GVHD mediated by human immune cells in vivo. Understanding the pathophysiology of this xenogeneic GVHD is therefore crucial for the design and interpretation of experiments performed with this model. In this article, we comprehensively review the cellular and molecular mechanisms governing GVHD in the most commonly used model of xenogeneic GVHD: PBMC-engrafted NOD/LtSz-PrkdcscidIL2rγtm1Wjl (NSG) mice. By re-analyzing public sequencing data, we also show that the clonal expansion and the transcriptional program of T cells in humanized mice closely reflect those in humans. Finally, we highlight the strengths and limitations of this model, as well as arguments in favor of its biological relevance for studying T-cell reactions against healthy tissues or cancer cells.