Abstract: Sugar content is the main attribute of a fruit's internal qualities. It is usually determined by soluble solids content (SSC). Since the traditional method used in detecting SSC with an Abbe-refractor is destructive, it is not suitable for on-line detection. To find a method to measure SSC nondestructively and quickly, the dielectric properties (relative dielectric constant and dielectric loss factor) of intact postharvest 99-1 nectarines at 1 day intervals during 10 days' storage were measured with a vector network analyzer (E5071C) and an open-ended coaxial-line probe (85070E) at 101 discrete frequencies over the frequency range of 20~4500 MHz. The sugar content of the fruit juice of each sample was measured with a digital refractometer. Altogether, 300 nectarines were used in the study. One-hundred and one relative dielectric constant values and 101 dielectric loss factor values at 101 discrete frequencies for each sample were used as variables to build models. A significance analysis was done to investigate whether frequencies and storage time had a significant influence on the values of permittivities. Sample set partitioning based on joint x-y distances (SPXY) was used to subset partitioning. Uninformative variables elimination (UVE) and successive projection algorithm (SPA) were applied to extract the characteristic variables from the original dielectric spectra of dielectric constant and dielectric loss factor. The modeling methods, such as partial least squares (PLS) and artificial neural network technology, such as support vector regression (SVR) and extreme learning machine (ELM) were applied to establish models for predicting SSC from permittivities. The experimental results showed that as the frequency increased, the relative dielectric constant of nectarines decreased, but the dielectric loss factor changed from decreasing to increasing. The analysis of variance indicated that storage time and frequency had a significant influence on dielectric properties. Based on SPXY, 243 samples were partitioned to a calibration set and 57 samples to a prediction set. One-hundred and eight variables were selected from 202 variables with UVE, and 14 characteristic variables were extracted with SPA. SPA was more effective than UVE in selecting useful information from the whole spectra of dielectric constant and dielectric loss factor and in simplifying a model for predicting SSC. Contrasted with PLS, SVR and ELM had better performance in predicting SSC under the SPA pretreatment method. SPA-ELM gave the highest correlation coefficient of predication set (0.887) and the lowest root mean square error of predication set (0.782). The study indicates that dielectric spectra combined SPA and artificial neural network technology could be applied in determining the sugar content of nectarines. It offers useful knowledge for developing nondestructive sensors for fruits' sugar content based on the frequency spectra of permittivities.