Distributed Denial-of-Service (DDoS) attack, using random spoofed source addresses, is popular because it can protect the attacker’s anonymity. It is very difficult to defent against this attack because it is very hard to differentiate bad traffic from the normal. In this paper, based on the source addresses distribution statistical feature, an effective defense scheme, which can differentiate vicious traffic from normal traffic, is presented. Based on a novel Extended Counting Bloom Filter (ECBF) data structure, this paper proposes an algorithm to identify normal addresses accurately. Once a normal address is sought out, packets from it will be forwarded with high priority, thus, normal traffic is protected. The simulation results show that this scheme can identify legitimate addresses accurately, protect legitimate traffic effectively, and give better protection to valuable long flows. Because the time complexity of the method is O(1), and it needs several MB memory space, it can be implemented in edge routers or network secure devices like firewalls to defend against random spoofed source address DDoS attacks.