In order to provide theoretical guidance for the grading of winter jujube maturity after harvesting, this study applied hyperspectral technique to obtain the characteristic wavelengths and calculate the spectral indices to achieve its maturity visualsort. A total of 336 samples of jujube with three types of maturity (immature fruit, white ripeness and primary red fruit, half red and red fruit) were collected and their hyperspectral information wereacquired. The samples were divided into training set (226) and testing set (110) using Kennard-Stone (K-S) method after the original spectral noise was reduced by Savitzky-Golay(S-G) smoothing algorithm. The characteristic wavelengths (CWs) were selected withsuccessive projections algorithm (SPA) and Competitive adaptive reweighted sampling (CARS). At the same time, 7 Spectral indices (SIs) were imported from the perspective of fruit varied physiological components. Three partial least squares discriminant analysis (PLS-DA) models were established based on the CWs selected by SPA and CARS and the introduced SIs, and the classification results of three models were compared. The results show that the discrimination accuracy of PLS-DA models based on two kind of CWs(selected by SPA and CARS, respectively)and SIs wereseparately 97.27%, 95.45%, and 98.18%. For the purpose of showing the discriminant results intuitively, a regression equation of the discriminant vector Y was fitted with SIs joint its PLS-DA regression coefficients, and the discriminant results were visually displayed by different colors in accordance with the rule that the corresponding category of the maximum value in Y is the sample belonging category. This study will contribute some proposals to visual grading of winter jujube maturity, and the imported SIs parameters will provide technical support for the manufacture of device that suitable for multiple fruits maturity sorting.