Abstract: The study of adulterants in food sampling data is very important for food safety early warning and risk prediction. In order to better explore the high-weight adulterants in food categories, we first design the food adulterant feature weight calculation model based on feature selection technology according to the idea of contrastive learning, and obtain sample classification results, feature information and common evaluation indicators in the model. Based on the above model, the visual analysis system for food adulterants is designed and implemented, which contains multiple linked views to help users understand the characteristics of food adulterants more intuitively and supports users to update the optimal feature combination through iterative interaction. Finally, the visual analysis system is used for the adulterant feature analysis of 89 202 unqualified samples of 24 food types nationwide from 2010 to 2020. The experimental results prove that our system can obtain the adulterant weights of food more conveniently and directly in an automated way, enhance the combination of adulterant features, and provide professionals with more comprehensive insight of food adulterants.