Abstract: This study first offered a new idea that tea comprehensive quality was discriminated with multi-sensor information fusion from near infrared spectroscopy and machine vision. In this experiment, four grades of roasted green tea were used as experimental targets. Principal component analysis (PCA) was implemented on these feature variables from image information and spectral information, and principal components (PCs) vectors were extracted as the inputs of pattern recognition. The discriminating model was built by BP neural network. The principal component factors from two-sensor information were optimized in building model. Experimental results showed that the optimal model was obtained under PCs=6 for image information and PCs=3 for spectral information. The discriminating rate equaled to 99% in training set, and 89% in prediction set. The overall results showed that it was feasible to discriminate tea comprehensive quality with two-sensor information fusion. The correct rate and robustness of the discriminating model from two-sensor information fusion were better than those of the model from the single-sensor information.