In view of the problem to effectively characterize the damage of dental caries, we explore a kind of polarization spectrum detection method which have nondestructive and low consumption characteristics, it’s a beneficial to supplement for the conventional detected methods such as chemical analysis, surface profilometry and microradiography. Based on the characteristic which polarization spectral sensitive to the observation sample surface microstructure, dental caries lesion due to the demineralization process, as result calcium and phosphate are dissolved from the enamel and dentin. The tooth surface microstructure has taken place different change; the structure present different degree of difference in light scattering properties and phase change. Considering different dental caries samples its surface microstructure changes strongly associated with the polarization information, we put forward a kind of method which can effectively characterize different teeth from caries by using polarization spectrum. Selected the 450, 550, 670 and 860 nm as four different observation research wavelengths, as well as selected six different dental caries samples, the degree of polarization as the parameter to describe different samples polarization characteristics. The experimental results show that the consistent observation waveband for different tooth samples is positively related to the degree of polarization parameters, as well as the same observation samples showed observation waveband is negatively related to the polarization characteristics. For further quantitative characterization, the relationship between polarization spectrum and tooth decay damage levels, the index correlation mathematical model which can interpret internal coupling was built. To effectively validate model robustness, it needed quantitative validation the model simulation results and measured data. We selected the quantitative evaluation factor such as the sum of squares due to error (SSE), root means squared error (RMSE), coefficient of determination (R-square). The results show that models R-square close to 1, as well as the SSE and RMSE values are small. The model characteristic of robustness and effectiveness was verified, which can effectively interpret different caries damage teeth samples from polarization coupling relation. This research is effectively extending the teeth caries detection method, as well as revealed polarization spectrum can be effective characterization and distinction to different dental caries samples. It is also developing a new kind of nondestructive and low-cost polarization spectrum detection technology.