To explore the potential of dielectric spectra in predicting internal qualities of pears, the dielectric constants and loss factors were measured by using openended coaxialline probe technology at 201 discrete frequencies from 20MHz to 4500MHz on 310 pears, picked from four different orchards, during 8week storage. Soluble solids content, firmness, and moisture content were considered as internal qualities. Sample set partitioning based on joint x-y distances was used to subset partitioning, and 233 samples were used in calibration set and 77 samples were used in prediction set. To simply establish model, successive projection algorithm method was applied to extract characteristic variables (CVs), and 15, 14 and 15 CVs were extracted for soluble solids content, firmness and moisture content, respectively. The modeling methods, such as least square support vector machine (LSSVM), extreme learning machine (ELM) and back propagation (BP) network were used to establish soluble solids content, firmness and moisture content determination models based on full dielectric spectra and extracted CVs by SPA. The results showed that the LSSVM model based on full dielectric spectra had the best soluble solids content determination performance and good prediction ability, with the correlation coefficient of calibration set of 0.974 and prediction set of 0.931, the rootmeansquare error of calibration set of 0592°Brix and prediction set of 0.868°Brix, and the highest residual prediction deviation of 2.65. The LSSVM model based on SPA could be used to predict the moisture content roughly. However, all models had poor prediction ability on firmness. The study indicates that dielectric spectra combined with LSSVM could be used to predict soluble solids content and moisture content of pears, but it is difficult to predict firmness using dielectric spectra. The study provides a method for nondestructive determination of soluble solids content and moisture content of pears.