Distributed denial of service attacks are the acts aiming at the exhaustion of the limited service resources within a target host and leading to the rejection of the valid user service request. During a DDoS attack, the target host is attacked by multiple, coordinated attack programs, often with disastrous results. Therefore, the effective detection, identification, treatment, and prevention of DDoS attacks are of great significance. Based on the research of DDoS attack principles, features and methods, combined with the possible scenarios of DDoS attacks, a Multi-Agent System-based DDoS attack detection method is proposed in this paper to implement DDoS attack detection for high-load communication scenarios. In this paper, we take the multi-layer communication protocols into consideration to carry out categorizing and analyzing DDoS attacks. Especially given the high-load communication scenarios, we make an effort to exploring a possible DDoS attack detection method with employing a target-driven multi-agent modeling methodology to detect DDoS attacks relying on considering the inherent characteristics of DDoS attacks. According to the experiments verification, the proposed DDoS attack detection method plays a better detection performance and is less relevant with the data unit granularity. Meanwhile, the method can effectively detect the target attacks after the sample training. The detection scheme based on the agent technology can reasonably perform the pre-set behaviors and with good scalability to meet the follow-further requirements of designing and implementing the prototype software.