Problem: The testing of head and neck squamous cell carcinoma (HNSCC) therapies will be facilitated by a tissue modeling system that incorporates ease of use while retaining the tri-dimensional architecture characteristic of these malignancies. We describe the phenotypic and gene expression differences in monolayer and tri-dimensional HNSCC culture systems. Methods: Normal oral epithelial cells (NOEC), primary HNSCC, HNSCC cell lines, and mixtures of NOEC and HNSCC were used to create modeled tri-dimensional tissue. Samples were stained with H&E or assessed for the expression of Ku, MMP2, and cytokeratins 13 and 19 via indirect immunofluorescence. Using Agilent cDNA Microarray analysis the gene expression profiles of monolayer and raft modeled tri-dimensional HNSCC cell lines were compared. Results: NOEC, HNSCC, and both together can be modeled with tri-dimensional differentiation and cytokeratin characteristics analogous to HNSCC in vivo tissue. HNSCC cell lines grown in rafts histologically resembled noninvasive carcinoma in situ; however, modeling of primary HNSCC resulted in a morphology of invasive HNSCC with areas of direct collagen invasion and MMP2 expression. The microarray cluster analysis and gene expression correlation results suggest that the individual cell lines themselves are the primary gene expression predictor and not the presence of tri-dimensional tissue architecture. There were only 5 of 12,814 genes with differential expression in paired analysis of raft versus monolayer HNSCC cultures. Conclusion: This tissue culture modeling system approximates the differentiation and tri-dimensional structure of in vivo NOEC, HNSCC, and both together on the same tissue plane. We have shown that raft modeled tri-dimensional HNSCC does not have a significantly different gene expression profile than the corresponding monolayer culture. Significance: The organotypic raft modeling system is an attractive model for translational investigation of future therapeutic interventions for HNSCC; however, the raft tri-dimensional modeling system may not be optimal for gene expression microarray-based experimental models of response to therapy or gene prediction analysis. Support: This paper has been submitted for the 2004 AAO Resident Research Award.