Successful drug development in oncology is grossly suboptimal, manifested by the very low percentage of new agents being developed that ultimately succeed in clinical approval. This poor success is in part due to the inability of standard cell-line xenograft models to accurately predict clinical success and to tailor chemotherapy specifically to a group of patients more likely to benefit from the therapy. Patient-derived xenografts (PDXs) maintain the histopathological architecture and molecular features of human tumors, and offer a potential solution to maximize drug development success and ultimately generate better outcomes for patients. Although imperfect in mimicking all aspects of human cancer, PDXs are a more predictable platform for preclinical evaluation of treatment effect and in selected cases can guide therapeutic decision making in the clinic. This article summarizes the current status of PDX models, challenges associated with modeling human cancer, and various approaches that have been applied to overcome these challenges and improve the clinical relevance of PDX cancer models.