Rapid and unified prediction of multiple soil properties is of great significance for regional soil resource surveys, soil quality assessment, and precision agricultural management. However, laboratory hyperspectral prediction of multiple soil properties is constrained by high spectral redundancy and the need for separate feature selection for individual soil properties, limiting the feasibility of unified prediction. This study aimed to construct a unified characteristic spectral band set to achieve rapid and simultaneous prediction of multiple soil properties using laboratory hyperspectral data. A total of 671 topsoil samples (0 – 20 cm) were collected from the major agricultural region of Northeast China, and their visible to near infrared hyperspectral reflectance spectra were measured under laboratory conditions. Mutual information (MI) was employed to identify characteristic spectral bands for soil organic carbon (SOC), pH, total nitrogen (TN), total phosphorus (TP), and total potassium (TK), and the variation in model performance under different MI thresholds was analyzed to determine the optimal characteristic band combination for each soil property. A characteristic spectral band set was constructed based on the frequency of the selected bands across different soil properties. Prediction performances of full spectral bands, property specific characteristic bands, and the characteristic spectral band set were compared using the random forest (RF) model. Laboratory hyperspectral data effectively predicted the major soil properties in Northeast China, although prediction performance varied among soil properties. SOC and TN achieved relatively high prediction accuracies, with R² values of 0.64 and 0.61, respectively, whereas the R² values for pH, TP, and TK were 0.43, 0.25, and 0.22, respectively. Feature selection significantly improved model performance, and an optimal MI threshold was identified for each soil property. Compared with the full spectral bands model, the prediction accuracies of SOC, pH, TN, TP, and TK increased by 8%, 19%, 12%, 22%, and 10%, respectively, using the optimal characteristic band combination. The sensitive spectral bands of different soil properties exhibited considerable overlap. Compared with the full spectral bands model, the characteristic spectral band set model improved the prediction accuracies of SOC, pH, TN, TP, and TK by 6%, 14%, 6%, 21%, and 6%, respectively. Feature selection based on laboratory hyperspectral data significantly improved the prediction accuracy of soil properties. The characteristic spectral band set effectively integrated the shared spectral information among multiple soil properties, enabling simultaneous prediction while maintaining satisfactory accuracy. This approach provides a promising strategy for multi-property soil modeling and rapid soil property assessment using laboratory hyperspectral spectroscopy.