The accuracy of the food safety evaluation model directly influences the accuracy of food safety situation assessment and forecast. Based on the hazard analysis critical control point theory (HACCP), a food safety evaluation index system was established from the perspective of the food supply chain. In order to detect the convergence speed and fitting degree of the model’s deviation, the analytic hierarchy process was utilized to improve the random initialization calculating weight method in the backward propagation neural network algorithm. Meanwhile, the sample data were trained and the test data were validated. The results showed that the BP neural network combined with AHP was high-precision, fast, and objective, which could be used to food safety evaluation of circulation links of production, processing, and sales.