Near infrared reflectance spectroscopy (NIRS) was evaluated as a tool to predict the chemical compositions of baled corn stalk silage. A total of 112 samples were used for determination of the pH value, crude protein (CP), crude ash (CA), dry matter (DM) and soluble carbohydrate (WSC). Samples were scanned with near infrared reflectance spectrometer and partial least-squares regression (PLSR) was used to predict the chemical compositions. The coefficients of determination of calibration (R2) and the coefficients of determination of validation (R2v) of the pH value, CP, CA and DM were higher than 0.85. The relative percent differences (RPD) of the pH value, CA and CP were higher than 2.5. The RPD of the WSC was higher than 2.0, and its R2V was 0.72. Therefore, the NIRS can provide accurate prediction of the pH value, CP and CA. It can also provide relatively accurate prediction of DM, but the variation range of DM should be enlarged. Moreover, it can be used for a rough estimate of WSC, but the precision should be improved.