Abstract:In order to test the prediction feasibility of total sugar, total nitrogen, nicotine, potassium and chlorine content in tobacco sheet, predictive models were established by partial least square (PLS) technique, based on the data of NIR diffuse reflective spectrum and determined chemical results of 250 tobacco sheet samples. Establishing parameters and the predictive effects of the models were evaluated in this study as well. Results showed that there was a remarkable correlation between predictive and chemical determinative values and all models present a good predictive veracity. The actual predictive mean relative error was 2.89% for total sugar model, and mean errors were 0.10%, 0.07%, 0.19% and 0.09% for total nitrogen, nicotine, potassium and chlorine model respectively. The relative standard deviations (RSD) were less than 4% for all the models. Accordingly, the analysis of total sugar, total nitrogen, nicotine, potassium and chlorine could be substituted by NIR spectroscopy, and could meet the need of rapid, convenient and mass analysis of the tobacco sheet samples.