Abstract: Objective: To establish a safety risk prediction model based on the analysis of food safety sampling data of vegetable in Jiangmen City from 2016 to 2020 and data mining. Methods: A total of 1945 samples of 10 kinds of vegetables from farmers' markets, wholesale markets, supermarkets and catering in Jiangmen City were collected and used to analyze the distribution of unqualified samples and unqualified items. Based on monitoring index and sample information, seven attributes including vegetable type, vegetable variety and monitoring place were selected as input and conclusion attribute was used as output. A risk analysis and prediction model of vegetable safety was established by back propagation (BP) neural network analysis. Results: The risk analysis showed that the qualified rates of sproutie vegetables, leafy vegetables, root vegetables and potato vegetables were 81.7%, 95.9% and 96.3%, respectively, lower than the averaged level. The excessive problems of sodium 4-chlorphenoxyacetate, chlorpyripyrix and lead were the main safety problems, 71.2% of the unqualified samples. A three-layer BP neural network model was constructed by data processing, optimal parameter screening and data training and validation, with an accuracy of 96.3%, a sensitivity of 96.8% and a specificity of 83.9%. Conclusion: The proposed model has good prediction performance, which can provide technical reference for food safety supervision. It is suggested that the multi-algorithm combination model can be built with the large data volume of rapid detection technology and BP neural network. Based on the the standardization of sample information registration, it is able to establish a risk analysis and prediction model with improved accuracy and broader application.