In the present work, 111 soil samples from 11 different Chinese apple plant areas were used to take the diffuse reflection spectra from 12 500 to 4 000 cm-1 by FT-NIR. The models of organic substance and pH value of soil samples were built by using partial least square regression (PLSR). The calibration model gave the correlation coefficients of 0.818 and 0.836 for the two values respectively, with the root mean square error of prediction (RMSEP) of 0.377 (%) and 0.251, respectively. In order to improve the robustness and performance of calibration, several spectra preprocessing methods were employed, including standard normalized variate (SNV), multiplicative scatter correction (MSC) and direct orthogonal signal correction (DOSC). Finally, the performance of DOSC was found to be the best for organic substance and pH value with RMSEP of 0.258(%) and 0.248, respectively. The results showed that the technology of NIR spectroscopy was useful to nondestructive determination of the organic substance and pH value of soil. These research findings provide theoretic base for fertilization and pomiculture by means of NIR diffuse reflection.