Wheat is not only the main grain in China, but also is an important feed and industrial raw material. Wheat is susceptible to scab, which can produce vomitoxin whose scientific name is Deoxynivalenol (DON). Vomitoxin is carcinogenic and pose a serious threat to human and animal health. In recent years, due to the frequent occurrence of extreme and abnormal weather, the risk of DON infection is on the rise, which has become the main factor affecting the quality and safety of wheat products. However, traditional methods for detecting DON content have obvious problems such as cumbersome and time-consuming detection process. Therefore, developing a fast, low-cost and online detection method is of great significance for the safe production and processing of wheat. Firstly, 200 wheat samples with different degrees of scab infection were collected from all parts of Jiangsu. After milling, the content of DON in wheat was determined by ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS), and then the visible/near-infrared spectral of wheat were collected online. The data processing steps are: pre-processing the spectrum by multi-scattering correction and second derivative, and extracting the characteristic wavelength according to the competitive adaptive reweighted sampling algorithm, then using linear discriminant analysis (LDA) and partial least squares discriminant analysis (PLS-DA) was used to establish a qualitative analysis model of wheat flour samples (with a national standard of 1 000 μg·kg-1), and a quantitative analysis model of DON content in wheat flour samples was established according to partial least squares regression (PLSR). UPLC-MS/MS results showed that the risk of wheat DON contamination was higher, and the over-standard rate of the tested samples was 50%. Visible/near-infrared spectroscopy analysis showed that the spectral characteristics of different DON content wheat samples had some differences. The original spectrum and the second derivative spectrum showed that the higher the DON content, the lower the absorbance at 1 420 nm. Due to the low absolute content of DON and the limited detection limit of spectroscopy, the obvious clustering trend could not be found by principal component analysis. However, the LDA and PLS-DA discriminant models constructed according to the full spectrum and the characteristic spectrum can quickly identify and screen sound and infection samples, and the best recognition rate was 87.69%. According to the quantitative analysis results, the PLSR model of DON content in wheat samples was not ideal. The optimal model results: the correlation coefficient (rp) of the prediction set was 0.688, the root mean square error (RMSEP) was 727 μg·kg-1, and the relative analysis deviation (RPD) was 1.38. The accuracy and robustness of the model needed to be further improved. It is feasible to use visible/near-infrared spectroscopy and chemometrics methods to achieve on-line discrimination and screening of wheat DON content exceeding the standard, which provides a technical reference for the rapid and quality detection of wheat products in China. However, the quantitative analysis of DON content needs further research to explore the influence of external factors on the model, and it is planned to expand the sample size, collect wheat samples from different regions and different varieties, and improve the accuracy and universality of the model.