Head and neck squamous cell carcinoma (HNSCC) reflect a highly heterogeneous and aggressive group of cancers for which concurrently many therapy options are associated with adverse side effects and resistance mechanisms. Although intense research is performed, the five-year survival rate is stagnating and remains low, with 50% on average. Preclinical drug testing is mainly based on animal testing, even though the gap between the species is known and the results are often misleading. Since tissue-engineered human-based models get more and more attention, due to their promising results in reflecting the human in vivo situation closely, models of normal oral mucosa (NOM) and tumor oral mucosa (TOM) have been developed in this thesis for advanced preclinical drug testing to improve HNSCC therapy. Hereby, NOM models reflected lining mucosa, with a defined basal membrane, the stratum basale, and stratum spinosum and primary tumor cells from patient-derived xenografts (PDX) and tumor cell lines could be integrated into the models by reflecting their original tumor grading-status. To elongate the model´s cultivation time, the commonly used collagen in the model´s matrix was replaced by a tight-knit web of esterified hyaluronic acid fibers, called Hyalograft 3D®. The development of the epithelium occurred slower but offered a continuous proliferation of up to 7 weeks in culture, in contrast to the 2 weeks limited functionality in the collagen-based models. This shows the high influence and importance of a well-defined extracellular matrix (ECM) for improved 3D-modeling. Drug effects have been investigated based on docetaxel and cetuximab, which are frequently used against head and neck squamous cell carcinoma, by comparing systemic and topical application routes. Docetaxel presented its potency by tumor mass reduction, with increased cell damage and inflammation as detected by lactate dehydrogenase and interleukin-6 release into the medium. Furthermore, a reduced proliferation (Ki-67), angiogenesis (HIF-1𝛼), and increased apoptosis (TUNEL) could be determined. Interestingly, the topical application often needed less docetaxel dosage to achieve the same cytostatic effects, compared to systemic application. In a first proof-of-concept study UHPLC-MS/MS analysis was integrated into the models, to enable automated sampling for docetaxel-concentrations inside the tumor tissue. Since sample preparations are dropped, this approach seems promising for future pharmacokinetic investigations. In contrast to docetaxel, cetuximab did not inhibit the proliferation of the tumor cells. Since cetuximab frequently triggers tumor resistances, it first had to be guaranteed, that the drug reaches its target site. Therefore, in cooperation with the physical institute of Freie Universität Berlin, the fluorescence-lifetime imaging microscopy and the atomic force microscopy-based infrared spectroscopy served for analysis. In summary, the established models could improve preclinical drug testing since the models closely reflect the human in vivo situation, are easily adaptable, and offer various drug-testings, be it based on morphology, pharmacokinetics, or drug detection. Future minimization of the models might allow high-throughput analysis and approaches for personalized medicine. Moreover, the integration of immune and blood cells could enable the study of a wider drug range and reflect the in vivo situation even more detailed. My developed and analyzed NOM and TOM models promise improved preclinical drug testing and promote the principles of 3R as the reduction, replacement, and refinement of animal testing.