Abstract: Traffic flow data is characterized by non-periodicity, nonlinearity and randomness. In order to accurately predict the traffic flow without ETC, it take measures to solve the traffic jam problem quickly and accurately, a traffic flow prediction system is proposed based on a neural network inference model. It is verified the prediction accuracy and adaptability of the ARIMA seasonal model, BP neural networks and RBF neural networks with various training functions by experiments. Relative to conventional forecasting method, the prediction method which is based on neural network is more adaptability and prediction accuracy is higher than conventional forecasting method.