In order to improve road traffic safety, an early warning algorithm based on Multi-Layer Perceptron Neural Network is proposed to solve the problem of low prediction accuracy caused by deterministic parameters in the current traffic safety early warning algorithm. The algorithm is based on artificial neural network (ANN). The relative distance, relative speed, driver′s driving style, the acceleration of preceding vehicle, the acceleration of following vehicle and the speed of following vehicle are used as the input of the system, and the warning level of traffic safety is the output of the system. The prediction value of traffic safety early warning level is obtained by training with sample data, and compared with the two early warning models of the traditional collision time algorithm and the stop distance algorithm. The experimental results show that the multilayer perceptron neural network early warning algorithm is superior to the traditional warning algorithm in the effectiveness and accuracy of early warning.