Advances in next-generation sequencing and bioinformatics have begun to reveal the complex genetic landscape in human cancer genomes, including oral squamous cell carcinoma (OSCC). Sophisticated preclinical models that fully represent intra- and inter-tumoral heterogeneity are required to understand the molecular diversity of cancer and achieve the goal of personalized therapies. Patient-derived xenograft (PDX) models generated from human tumor samples that can retain the histological and genetic features of their donor tumors have been shown to be the preferred preclinical tool in translational cancer research compared with other conventional preclinical models. Specifically, genetically well-defined PDX models can be applied to accelerate targeted antitumor drug development and biomarker discovery. Recently, we have successfully established and characterized an OSCC PDX panel as part of our tumor bio-bank for translational cancer research. In this paper, we discuss the establishment, characterization, and preclinical applications of the PDX models. In particular, we focus on the classification and applications of the PDX models based on validated annotations, including clinicopathological features, genomic profiles, and pharmacological testing information. We also explore the translational value of this well-annotated PDX panel in the development of co-clinical trials for patient stratification and treatment optimization in the near future. Although various limitations still exist, this preclinical approach should be further tested and improved.