In order to detect wheat quality rapidly and nondestructively, NIR wheat quality quick detection system was developed on the base of grating technology. To test accuracy, repeatability and stability of this self-made system, Bruker MPA spectroscopy was selected as target analyzer and 56 wheat samples were analyzed by building and validating PLS calibration models. In the 4 models of the self-made system, the coefficient of determination R2 is 92.38%, 93.48%, 93.16% and 94.44%; root mean square error of cross validation RMSECV=0.405, 0.374, 0.383, 0.346; ratio of performance to standard deviate RPD=3.62, 3.39, 3.82, 4.24, respectively. And evaluating indicators of validating results in the 4 models are as follows: R2=96.97%, 94.22%, 96.62% and 96.34%; Root mean square error of prediction RMSEP=0.221, 0.305, 0.233 and 0.243 respectively. The model of MPA spectroscopy gave an R2 of 95.99%, a RMSECV of 0.293, RPD of 5 and validation results are R2 of 98.31%, RMSEP of 0.165, respectively. The results show that the models of self-made instrument have good prediction performance, stability and repeatability, and wavelength and absorbance of the obtained spectra have a good repeatability. The prediction effect of single spectrum is not ideal, but it can be improved by using average spectrum of repeated acquisition. NIR wheat quality quick detection system can detect wheat quality with good performance.