Jujube was rich in nutrition and variety. In different varieties, there were very different from the market price to the qualities of internal and external. In order to realize the rapid and non-destructive detection of fresh jujubes' classification, Ban jujube, Jixin jujube and Xiang jujube were selected as research objects to collect their visible/near infrared spectral data dynamically. A combination of Moving Smoothing and Multiplicative Scatter Correction (MSC) was applied as the pretreatment method. After the pretreatment, the characteristic wavelengths extracted by Successive Projections Algorithm (SPA) were 980 nm, 1860, 1341, 1386, 2096, 1831, 1910, 1628, 441, 768 and 601 nm, respectively. And the importance reduced in accordance with the order. The 11 characteristic wavelengths were adopted as input variable to established Extreme Learning Machine (ELM) classification model, which was used for prediction. Comparing the ELM model's classification accuracy with other methods' classification accuracy such as Partial Least Squares Discriminant Analysis (PLS-DA) and Least Squares Support Vecor Machines (LS-SVM), the result indicated that: the R2 and the RMSEC of the SPA-ELM model was 0.97238 and 0.018724, respectively. The classification accuracy of the SPA-ELM model was 100% as good as the SPA-PLS-DA and SPA-LS-SVM. ELM was an effective classification method. This study provides a new theoretical basis for detection of fresh jujubes' classification.