In this study, the feasibility of the rapid non-destructive testing method for the pomegranate quality in the Sichuan and Yunnan Provinces is investigated based on the visible/near-infrared diffuse transmission spectroscopy technique. First, the near-infrared spectra of pomegranates are obtained using a dynamic online detection device which can effectively suppress the effect of stray light, and the actual sugar content value is measured. In combination with the principal component analysis method, the cluster analysis of pomegranates from different producing areas can approximately divide the samples into two categories. Further, a partial least squares discrimination analysis model is developed for the pomegranates from two distinct producing areas, which exhibits an accuracy greater than 97%. Meanwhile, multiple pretreatment methods, such as Savitzky-Golay smoothing, normalization, baseline correction, and multiplicative signal correction, are employed to establish a single model for two pomegranate types. Based on the obtained results, the baseline correction method is observed to be better than the other examined methods. In particular, the correlation coefficient of the prediction set (R_p) of the established Sichuan pomegranate model is 0.82, the root mean square error of prediction set (RMSEP) is 0.37, the correlation coefficient of the calibration set (R_c) is 0.90, and the root mean square error of calibration set (RMSEC) is 0.31. However, for the Yunnan pomegranate model, the R_p is 0.81, the RMSEP is 0.33, the R_c is 0.87, and the RMSEC is 0.27. In the post-sorting verification experiment for samples not involved in modeling, the discriminating rate of pomegranates in both the producing areas is 95%, whereas the sugar content sorting accuracy is 92.5%. Thus, the near-infrared spectroscopy is of considerable significance with respect to the discrimination of the pomegranate producing area and the sorting of its sugar content and may form the basis for future pomegranate online sorting research.