Abstract: In order to determine the cellulose in tobacco rapidly, a predicting model of cellulose in tobacco leaves, based on the data of NIR diffuse reflective spectrum and chemically determined cellulose content of 1214 tobacco samples, was established with the partial least square (PLS) statistic method.The parameters and predictive effects of the model were evaluated, the determination coefficient of the model was 0.9649, the mean relative deviation between predicted values and measurements was less than 2%.The model is suitable for the occasion where a large number of tobacco samples are to be analyzed in short time.