Near infrared (NIR) spectroscopy has been widely studied and used for rapid and nondestructive measurement of internal qualities of fruits such as sugar content, acidity, firmness, etc. The objective of the present research was to study the potential of NIR diffuse reflectance spectroscopy as a nondestructive method for the determination of sugar content of Jinhua peaches. NIR spectral data were acquired in the spectral region between 800 nm and 2500 nm using a FT-NIR spectrometer with a bifurcated optic fiber and an InGaAs detector. Statistical models were developed using partial least square regression (PLSR) method by TQ Analyst software. The results of PLSR models for peach flesh of different parts and juice indicated that the model based on average spectra of nine measurements in three different parts of each fruit and the corresponding sugar content obtained better results than those models based on the flesh of one part of each fruit (three measurements) or juice. Spectral data preprocessing of derivative and scattering correction was also discussed. The results showed that the models based on original spectra were better than those based on derivative spectra; and spectra scattering correction could improve the performance of PLSR models. Finally, two models were established based on spectra after multiplicative scattering correction (MSC) and standard normal variate (SNV) preprocessing. The correlation coefficients of calibration and leave-one-out cross-validation of the two models were the same, Rcal = 0.997 and Rcross-v = 0.939. These results show that it is feasible to use NIR spectroscopy technique for quantitative analysis of peach sugar content.