Near infrared spectroscopy (NIR) is an instrumental method applied for rapidly measuring the NIR spectra of pulverized tobacco samples and computing the chemical compositions from the spectral data. In the present paper, a mixed algorithm was employed for building the nonlinear model of NIR and the total sugar of tobacco samples. The mixed algorithm was combined with Partial Least Squares (PLS) method and Artificial Neural Network (ANN). The model based on the mixed algorithm was divided into two parts: linear part and nonlinear part, and the corresponding model of each part was built respectively. Compared with the classical multivariate calibration methods such as Principle Component Regression (PCR), PLS and nonlinear PLS (NPLS), the proposed procedure performed much better. The results showed that the mixed algorithm could be used for the quantitative analysis of the total sugar in tobacco samples.