During precancer development in epithelium, neoplastic cells remodel the underlying stroma, for example, the basement membrane, capillaries, fibroblasts, and extracellular matrix. The purpose of this study is to investigate the relationship between the nonlinear optical signals from the collagen matrix in stroma and the progression of early epithelial carcinogenesis. Two-photon excited fluorescence (TPEF) and second harmonic generation (SHG) signals were measured from the stroma of hamster oral cheek pouch in vivo. We found that three features, including the intensity ratio of TPEF over SHG, the spatial frequency distribution, and the texture feature of SHG images, provide the quantitative identification of epithelial precancer at different pathologic stages. We demonstrated that the combination of all three features by using a support vector machine algorithm can significantly improve the accuracy in the detection of epithelial precancer.