Pulsing Denial-of-Service (PDoS) attacks seriously degrade the throughput of TCP flows and consequently pose a grave detrimental effect on network performance. The fact that they generate less traffic than traditional flood-based attacks makes PDoS detection more difficult. Most of the conventional PDoS detection shemes focus on the periodical pattern of the pulse trains. Therefore, attackers can easily escape the detection system by merely controlling the timing of pulse transmission. In this paper, we propose a novel and robust PDoS detection method which capitalizes on the bandwidth usage condition of network traffic in distinguishing the congestion due to normal traffic from that due to PDoS attacks. Simulation experiments have demonstrated the effectiveness of the proposed scheme in detecting PDoS attacks.