Near infrared (NIR) spectroscopy was used to identify the origins of two representative of Shandong green tea (Laoshan green tea and Rizhao green tea) rapidly and non-destructively. Several preprocessing methods of NIR, such as smoothing, first-and second-derivative were compared. Moving window back propagation artificial neural network (MW-BP-ANN) was used to select characteristic spectral variables. It was found that the first-derivative and MW-BP-ANN processing techniques improved the predictive abilities of the support vector machine (SVM) classification models. The best estimated identification accuracy can be improved to 98.33%, which demonstrates that the spectral variables selection method is significant for the predictive ability of origin identification models.