Objective To establish a prediction model for the soluble solid content (SSC) and firmness of prunes in Xinjiang using visible and near-infrared (Vis-NIR) spectroscopy, enabling nondestructive and accurate evaluation of the internal quality of the fruit.Methods Taking 280 prunes in Xinjiang as research samples, the study collects Vis-NIR reflectance spectra at different storage periods with a near-infrared system. Principal component analysis-Mahalanobis distance (PCA-MD) is applied to eliminate abnormal samples. Then, the remaining dataset is divided into a calibration set and a prediction set at a ratio of 3∶1 using the Kennard-Stone (K-S) algorithm. Partial least squares (PLS) and support vector regression (SVR) models are employed to compare the effects of different preprocessing methods on the raw spectra, including Savitzky-Golay smoothing (SG), standard normal variable transform (SNV), first-order derivative (1st-D), and normalization (Norm). Characteristic wavelengths of the prunes' spectra are selected using competitive adaptive reweighted sampling (CARS), bootstrapping soft shrinkage (BOSS), and interval combination optimization (ICO). Based on the selected features, regression models for the quality prediction of prunes are established using PLS and SVR algorithms.Results For the SSC prediction of prunes in Xinjiang, the SNV-CARS-SVR model demonstrates the best performance, with a determination coefficient (Rp2) of 0.866, a root mean square error of prediction (RMSEP) of 0.651, and a residual predictive deviation (RPD) of 1.956. For firmness prediction, the Norm-BOSS-PLS model achieves the best results, with an Rp2 of 0.894, an RMSEP of 0.740, and a RPD of 2.207.Conclusion The nondestructive prediction of the SSC and firmness of prunes in Xinjiang using Vis-NIR spectroscopy is demonstrated to be feasible and holds considerable potential for practical application.