Totally 180 samples coming from one orchard were divided into calibration set with 150 samples and prediction set with 30 samples. Field Spec3 spectrometer was used for collecting spectra data of 180 fresh jujube samples separately. Then successive projection algorithm and stepwise regression analysis combined with spectral theory were used to process the spectral data after MSC pretreatment. Characteristic wavelengths of 150 samples in calibration set were selected by using SPA and SRA, and the partial least square (PLS) and LS—SVM methods were used to establish models of the fresh jujube soluble solids with the whole spectrum and characteristic wavelengths selected by using SPA and SRA. At last, the models of MSC—PLS model, MSC—LS—SVM model, the MSC—SPA—PLS model, the MSC—SPA—LS—SVM model, the MSC—SRA—PLS model and the MSC—SRA—LS—SVM model were used to predict the soluble solids of 30 samples in the prediction set. The results showed that the correlation coefficient and the root mean square error of prediction of MSC—PLS model for full band are 0.8874 and 1.0889 and that is the best. The correlation coefficient and the root mean square error of prediction of MSC—SPA—PLS model and MSC—SRA—PLS model are 0.7990, 1.4078 and 0.8224, 1.3851, and they are less precise than the MSC—PLS model. The correlation coefficient and the root mean square error of prediction of MSC—SPA—LS—SVM model are 0.7963 and 1.1458, and that is more precise than the MSC—LS—SVM model. The precision of MSC—SRA—LS—SVM model is very low and is not suitable.