Near infrared spectra of 138 tobacco leaves samples of dark sun–cured tobacco were measured by fourier transform near infrared diffuse reflectance spectroscopy. Seven outliers were removed with chemometrics method and the remaining 131 samples were divided into training set and prediction set by the Kennard–Stone algorithm. Spectrum values and routine chemical indexes determined by chemometrics method were combined to establish calibration models by partial least squares method. The accuracy of the models was improved by the spectral preprocessing and variable selection methods. Quantitative analysis of total chlorine, total nitrogen, total alkaloid, total potassium, total sugar, reducing sugar and protein of dark sun–cured tobacco was achieved and the results obtained by infrared method have good correlation with those by the conventional methods, with correlation coefficients of all the models bigger than 0.90, indicating the calibration models can be used for rapid determination of the routine chemical indexes of dark sun–cured tobacco.