A study on in vivo measurement of autofluorescence spectra for hamster buccal pouch and development of oral carcinogenesis identification algorithm is presented. The measurement was preceded with a fiber-optics based fluorescence spectroscopy system. In total 75 samples, including 14 hyperkratosis, 23 normal, 28 dysplasia, and 10 SCC, were separated into 4 categories. All the spectra were normalized to have the same area below the spectrum curve. The results show that the autofluorescence spectra start to change as soon as the tissues have morphological alternation (eg hyperkratosis). The differences of ratios between the areas under 380+/- 15 nm and 460+/- 15 nm (denoted as A380+/- 15/ A460+/- 15) among categories are statistically significant. To develop a diagnostic algorithm for early neoplasia detection and evaluate its performance, a PLS discriminant analysis with cross-validation technique was proceeded. Sample points on the PLS score plot were grouped as four categories. By selecting suitable threshold, the accuracy rates for classifying 4 categories of samples are 86% (hyperkratosis), 87% (normal), 90% (dysplasia), and 100% (SCC), respectively. The results reveal that the autofluorescence spectroscopy technique is potential for in vivo detection of early neoplasia of oral tissues.