Abstract: Diffuse reflectance spectrum of near infrared (NIR) ranging from 1300 nm to 2100 nm was investigated for its feasibility to determine the sugar content (SC) of apple non-destructively. For each sampled Fuji apple before and after being peeled, the mean spectrum was calculated by averaging the spectra collected at four sampling locations around its equator, and the value of SC was measured with a Brix refractometer after the whole apple being squeezed into juice. Then all data were analyzed by principal component regression (PCR) and partial least square (PLS). Results show that NIR diffuse reflectance between 1300 nm and 2100 nm is applicable to non-destructive determination of the SCs of apples, and that the PLS could produce a better prediction effect than PCR. Models for predicting SC of whole apple developed from spectra of single measured sampling locations and from mean-spectrum of four spectra collected on each apple were studied. Result indicates that the latter shows better prediction precision than the former. So, it is suggested that mean-spectrum of several spectra collected by several fibers around the equator of each apple should be a better solution to predict SC on-line.