The content of sugar is an important quality index for pears. However, the traditional sugar measurement methods are time-consuming and destructive. In the present study, the authors measured the sugar content of pears using visible and near infrared diffuse reflection spectroscopy. The pretreatment methods of multiplicative scatter correction (MSC), baseline correction, standard normal variate (SNV) transformation, and moving average algorithms were used on the original absorbance spectrum. Results indicate that the absorbance spectra after pretreatment are better than the original absorbance spectra for prediction. Partial least squares (PLS) regression was also used on the original absorbance spectrum and the absorbance spectrum after moving average and baseline correction. It follows that the forecast accuracy of the absorbance spectra after moving average is higher than that of the original absorbance spectra. The models gave good predictions of the sugar content of pears, with corresponding r values of 0.990 8, and standard errors of predictions of 0.019 0.