Non-destructive inspection of the interior and exterior quality of fruit has always been a research topic because many subjective assessing methods limited to the exterior measurements with poor repeatability and tedious procedures are still widely used. In this study, a hyperspectral-imaging technique was developed to realize a fast, accurate and objective grading of Chinese pears. The morphological features and spectral responses on sugar and water content can be extracted simultaneously. The feature wavelengths for water content prediction(462, 502, 592, 706 and 957 nm) and for sugar content prediction(500, 703, 816, 875 and 920 nm) were selected based on partial least squares analysis. Artificial Neural Network was engaged to establish the prediction model for the water and sugar contents. The results show that the ANN model could predict water and sugar contents of pear samples with correlation coefficient of 0.996 and 0.94, respectively. RMSEP was 4.24% for water content and 0.5°Brix for sugar content. For weight prediction, the correlation coefficient between predicted and real weight was 0.93.