In the present paper, 60 whole wheat flour samples were prepared and corresponding NIR spectra were collected. After the determinations of calibration range, several spectral sub-regions containing calibration range and prepared for the correction using multiple scattering correction (MSC) were obtained in the whole spectral region firstly, and MSC correction based on different spectral sub-region followed subsequently. Corresponding spectral data were obtained from the fixed calibration range of the spectra corrected based on different regions. Several partial least squares regression (PLSR) calibration models for analyzing protein content of whole wheat flour were established based on corresponding spectral data, and according to the performance about the calibration coefficient and the root mean square error of cross validation (RMSEV) of every calibration, the effects of MSC pretreatment spectral region on PLSR calibration results were investigated and the pretreatment spectral regions were optimized by comparing the performance of more calibration models. For the optimized calibration, the calibration coefficient and the RMSECV improved compared with the calibration established based on the spectral data corrected using MSC in the fixed region of calibration. The correlation coefficient can be raised from 0.96 to 0.98 and RMSECV can be decreased from 0.37% to 0.32%. The results show that the capability of MSC on correcting the spectral interference information of non-chemical absorption can be influenced by preprocessing spectral regions, the performance of calibration model can be improved by optimizing the MSC pretreatment spectral region, and the appropriate pretreatment spectral region is prerequisite to obtain the best calibration results while using MSC for near-infrared spectra analysis.