Abstract: Urban traffic system is very complicated.For the complexity of traffic flow and the real－time requirement of traffic control,Reinforcement Learning can be applied to the traffic signal control problem.It makes dynamic decisions depending on the real－time traffic information,and adapts to the environment automatically so as to achieve better results.As the state space is too huge to set up a linear table,this paper proposes a hybrid approach which is based on Reinforcement Learning and Artificial Neural Network to solve the traffic signal control problem of multi intersection.By comparing this approach with traditional fix sequence traffic signal control strategy in simulative environment,the result indicates that the new approach is better.