Soluble solids content is an important index to evaluate apple quality. The works aims to provide theoreti-cal basis for the development of detection equipment for rapidly predicting apple quality. The hyperspectral image acquisition system was used to collect hyperspectral images of "Fuji" apples and obtain the reflectance spectra in the regions of interest. The successive projection algorithm was used for the dimensionality reduction of the reflectance spectra subject to standard normal variation preprocessing. The multi linear regression model was established based on selected characteristic wavelengths to predict soluble solids content of apples. The results showed that 12 wavelengths as characteristic spectra were extracted by successive projection algorithm from 256 full spectra, and the working efficiency of multi linear regression prediction model was obviously improved. The multi linear regression model based on characteristic spectra had better calibration ability (RC=0.804, RCm=0.665%) and prediction ability (RP=0.859, RPm=0.413%). The prediction model established in this study for detection of soluble solids content of apples has stable properties and can meet the requirements of practical application.