Harvest maturity is an important factor that can affect plum quality during the storage period. In order to achieve rapid non-destructive determination of plum maturity level, in this study, 163 plum samples were classified into three maturity levels-early stage (n = 53), middle stage (n = 55), and late stage (n = 55)-according to different fruit development times after flowering. Near infrared spectroscopy (NIR) was used to classify plum samples into different maturity levels. The accuracies of the classification models established by Mahalanobis distances analysis, soft independent modeling of class analogy, and least squares-support vector machine were compared. Mahalanobis distance model with the first 20 principal components (PCs) extracted from original spectra as inputs yielded the best result, and the correct classification rates for calibration set and prediction set were 96.33% and 96.30%, respectively. Soluble solid content (SSC), titratable acid, and firmness were analyzed using single factor one-way analysis of variance (ANOVA). There were significant differences in each index, and the greatest difference was found in SSC. The PCs that were extracted from the data of the three quality indices had very similar clustering tendencies as those extracted from spectrum data. These results showed that combining NIR with chemometrics to discriminate plums of different maturity levels was feasible and effective, and the difference in quality indices could be a physicochemical validation for the classification results from NIR.