The detection of dysplasia and early cancer is important because of the improved survival rates associated with early treatment of cancer. Raman spectroscopy is sensitive to the changes in molecular composition and molecular conformation that occur in tissue during carcinogenesis, and recent developments in fiber-optic probe technology enable its application as an in vivo technique. In this study, the potential of Raman spectroscopy for in vivo classification of normal and dysplastic tissue was investigated. A rat model was used for this purpose, in which dysplasia in the epithelium of the palate was induced by topical application of the carcinogen 4-nitroquinoline 1-oxide. High quality in vivo spectra of normal and dysplastic rat palate tissue, obtained using signal integration times of 100 s were used to create tissue classification models based on multivariate statistical analysis methods. These were tested with an independent set of in vivo spectra, obtained using signal collection times of 10 s. The best performing model, in which signal variance due to signal contributions of the palatal bone was eliminated, was able to distinguish between normal tissue, low-grade dysplasia, and high-grade dysplasia/carcinoma in situ with a selectivity of 0.93 and a sensitivity of 0.78 for detecting low-grade dysplasia and a specificity of 1 and a sensitivity of 1 for detecting high-grade dysplasia/ carcinoma in situ.