Objective To analyze the co-occurrence characteristics of mycotoxins in wheat， a classification model based on climatic regions of China was built.Methods A total of 887 wheat samples collected from 12 provinces/autonomous regions were analyzed for the concentrations of eight mycotoxins， including deoxynivalenol， nivalenol， aflatoxins， ochratoxin A， fumonisins， zearalenone， T-2 and HT-2. All the samples were divided into three groups， temperate continental climate， temperate monsoon climate and subtropical monsoon climate， according to the climate types of their sampling sites. The borderline-SMOTE method was used for sample augment to balance the data set. Principal component analysis （PCA） was applied for data dimension reduction， and the first two dimensions with a cumulative contribution rate of 97% were chosen as the characteristics of the original data. The classification of the data feature was implemented using the k-nearest neighbor （KNN） nonlinear classifier， and the parameters of the KNN model were optimized using GridSearchCV. Confusion Matrix， accuracy， recall rate and F1 score were used as the indexes for model evaluation， and the performance of this model was compared with three other common models， including support vector machine， random forest and artificial neural network.Results The classification accuracy of eight mycotoxins in wheat using the combination of borderline-SMOTE， PCA and KNN model reached 98.31%， and the performance of this approach was superior to other frequently used methods.Conclusion The classification model established in this paper can effectively categorize the wheat samples into three climate regions based on the co-occurrence characteristics of mycotoxins， which provides a basis for region-specific cumulative risk assessment of combined mycotoxin exposure and puts forward a food classification method based on multiple food safety indicators.