Abstract: The initial weights and thresholds of back propagation ( BP) neural network were optimized by genetic algorithm ( GA) to accelerate the network convergence and improve the prediction precision.The liquid milk in dairy products was used as the experimental material to establish the safety evaluation index system. The GA-BP neural network was used as the evaluation model to fit the daily data of liquid milk.The convergence rate and the fitting degree of the model were verified by the test data.The results showed that GA-BP was more stable than BP neural network and could converge quickly, and the simulation error of GA-BP neural network was smaller.When the number of nodes was 9, GA-BP neural network had the best fitting effect to liquid milk, and the prediction precision was high. So GA-BP neural network was a feasible method to evaluate the safety of liquid milk.