Abstract: In order to explore the coordination of tobacco chemical compositions (shmuck value, ratio of sugar and nicotine, ratio of nitrogen and nicotine), the data of near infrared spectrum were pretreated by different methods. The NIR spectral detection models were established based on the effective wavelengths. The regression model was checked by PLS to cross training set. The regression model of shmuck value was established after median filter smoothing of full spectra with coefficient r=0.9861, RMSE=0.0548 in prediction set. The regression model of the ratio of sugar and nicotine was established after convolution smoothing and 2nd derivative of full spectra with coefficient r=0.9498, RMSE=0.9095 in prediction set. 30 effective wavelength data in processing array were selected to set up the regression model of the ratio of nitrogen and nicotine based on the better Norris first derivative array, which had the regression coefficient r=0.9202, RMSE=0.6947. The results indicated that the NIR spectra could better predict the shmuck value, ratio of sugar and nicotine and ratio of nitrogen and nicotine of tobacco.