The models of quantitative analysis of brix and pol in sugar cane juice were established by using near infrared spectroscopy (NIR) coupled with the back propagation-artifical neural network method (BP-ANN). The spectra of cane juice samples were obtained by the way of 2 mm optical length transmission and using the NIR spectrometer of 1 000-1 800 nm wavelength. Firstly, the data of original spectra were pretreated by Savitzky-Golay derivative and mean-centering. Secondly, the wavelength range of model was optimized by using correlation coefficient method coupled with the characteristic absorbance of the spectrum. Finally, the principal components, obtained by PLS dimension-reducing, were inputed into BP-ANN. The calibration models were established by calibration set and validated by prediction set. The results showed that the related coefficients (R2) of prediction for brix and pol were 0.982 and 0.979, respectively; and the standard errors of prediction (SEP) for brix and pol were 0.159 and 0.137, respectively. BP-ANN was more accurate in the predition of brix and pol compared with the partial least square method (PLS). The method can be applied to fast and accurate determintaton of brix and pol in sugar cane juice.