A non-destructive method for on-line determining the internal comprehensive quality of Huping jujube fruit was investigated based on visible/near-infrared reflection spectrum. Moisture content, soluble solid content, firmness, soluble protein content and vitamin C content were respectively used as internal indexes to assess the quality of Huping jujube at full ripe stage. Competitive adaptive reweighted sampling (CARS) was applied to select sensitive wavelengths. Models of the least squares-support vector machines (LS-SVM) were built based on the sensitive wavelengths respectively. The model of firmness showed that the correlation coefficient of prediction was 0.9452 and root mean square error of prediction was 41.6849. The other four models obtained the better results with the correlation coefficient of each prediction over 0.9230 and root mean square error of each prediction from 0.2674 to 3.7792. Then, the correlation was analyzed between the quality indexes. The results indicated that an extremely significant or a significant correlation was revealed between any two indexes in the P<0.01 or P<0.05 levels. Therefore, factor analysis was carried out on five internal quality index of fresh jujube to develop the internal comprehensive quality index, and the CARS-LS-SVM model of this index was established. The results indicated that the correlation coefficient of prediction was 0.9241 and root mean square error of prediction was 6.0635. This research showed that the established CARS-LS-SVM model was effective to realize evaluation of the internal comprehensive quality on fresh jujube. This research provided theoretical basis for on-line, rapid and non-destructive detection on internal comprehensive quality of fresh jujube.