Abstract: An effective NIR spectral detection model of tobacco total nitrogen content was explored, and effective wavelengths for the detection of tobacco total nitrogen content between 1100~2500nm were found. A variety of spectroscopic processing methods were used to deal with the original spectrum. The method of first derivative was better than others. Partial least squares regression (PLS) was established the regression model, effective wavelengths were selected by Martens’ uncertainty test. Regression model based on all wavelengths had the regression coefficient r==0.9930，RMSE=0.0490 in training set, r=0.9708，RMSE=0.0996 in cross validation set, and r=0.9747，RMSE=0.0884 in prediction set. Regression model based on effective wavelengths had the regression coefficient r==0.9937，RMSE=0.0464 in training set, r=0.9744,RMSE=0.0938 in cross validation set, and r=0.9610,RMSE=0.1116 in prediction set. The absolute error of prediction value and chemical value was less than 0.227%, and the relative error is not more than 0.1%. It showed that using NIR spectroscopy to detect tobacco total nitrogen content was good. It was feasible to select effective wavelengths using Martens’ uncertainty test and to detect tobacco total nitrogen content using these effective wavelengths.