To resist the risk of amplification DDoS attack, which is likely to cause significant damage to cyber security, a reinforcement learning method is proposed. Taking DNS as the target of attack, a simplified amplification attack model is constructed. The transition probability between different states is obtained by using the model-free method. Then, the reinforcement learning method is used to build up to defense the attack, and the traffic suppression strategy is formulated by learning the amplification attack mode. Finally, the simulation results show that the proposed reinforcement learning method can effectively dig out the traffic pattern of amplification DDoS attack and intelligently resist the threat.