Partial least squares regression (PLSR) calibration model will be effect by the wavelength change of a single instrument at a different time and the wavelength consistency of multiple instruments. The process of near-infrared spectroscopy analysis, this problems can be unified as the effect of wavelength drift on chemometric calibration model. In this paper, taking the analysis of crude protein in wheat flour as an example, two calibration models I and II were established by partial least squares regression (PLSR) method within different spectral regions. Different types and amplitudes of wavelength shift information were generated by computer and superimposed into the spectra of the validation set to produce wavelength shift information relative to the spectra of calibration set, the effect of wavelength drift on PLSR calibration model was studied by adding different types and amplitude of wavelength drift information to the spectra of the validation set samples. The results show that the RMSEP of every model is no more than 0.3% and the corresponding Rp is no less than 0.98 when there is no wavelength drift information in the spectra of validation set samples. When the wavelength drift at different wavelengths is constant, the RMSEP increases as the wavelength drift amplitude increases, the RMSEP increases to 3.69% when the wavelength drift is -32 cm^-1, and the Rp is almost constant; When the wavelength drift varies randomly at different wavelengths, the prediction results of model II based on long wavelength regions are almost not affected. The model II is corrected by a series of spectra added to the calibration set with different wavelength drift information, the RMSEP of the corrected model is 0.3%, The influence of wavelength drift information on RMSEP has been almost eliminated, but the number of regression factors used to establish corredted model increases significantly from 3 to 8, the robustness of the model varies greatly. In general, the RMSEP can be polished by correcting the prediction results to ensure the accuracy of the analysis results if the amplitude of wavelength drift is slight. This study provides an experimental basis for determining the design instrument parameters and operating procedures to improve the reliability of NIR analysis results.