Background: Oropharyngeal squamous cell carcinomas (OPSCC) are rising rapidly in incidence due to Human Papillomavirus (HPV) and/or tobacco smoking. Prognosis is better for patients with HPV-positive disease, but may also be influenced by tobacco smoking and other factors. There is a need to individualize treatment to minimize morbidity and improve prognosis. Patient-derived xenografts (PDX) is an emerging pre-clinical research model that may more accurately reflect the human disease, and is an attractive platform to study disease biology and develop treatments and biomarkers. In this study we describe the establishment of PDX models, compare PDX tumors to the human original, and assess the suitability of this model for radiotherapy research and biomarker development. Material and methods: Tumor biopsies from 34 patients with previously untreated OPSCC were implanted in immunodeficient mice, giving rise to 12 squamous cell carcinoma PDX models (7 HPV+, 5 HPV-). Primary and PDX tumors were characterized extensively, examining histology, immunohistochemistry, cancer gene sequencing and gene expression analysis. Radiosensitivity was assessed in vivo in a growth delay assay. Results: Established PDX models maintained histological and immunohistochemical characteristics as well as HPV-status of the primary tumor. Important cancer driver gene mutations, e.g., in TP53, PIK3CA and others, were preserved. Gene expression related to cancer stem cell markers and gene expression subtype were preserved, while gene expression related to hypoxia and immune response differed. Radiosensitivity studies showed high concordance with clinical observations. Conclusion: PDX from OPSCC preserves important molecular characteristics of the human primary tumor. Radiosensitivity were in accordance with clinically observed treatment response. The PDX model is a clinically relevant surrogate model of head and neck cancer. Perspectives include increased understanding of disease biology, which could lead to development of novel treatments and biomarkers.