In order to detect the increasingly serious distributed denial of service (DDoS) attack on the internet, an algorithm for detecting DDoS attack based on multi-dimensional information entropy is proposed. First of all, according to the property of DDoS attack, the multi-dimensional detecting vector which is capable of distinguishing attack from normal traffic is constructed based on conditional entropy and discrepant entropy. Then the sliding multi-dimensional non-parameter CUSUM algorithm with the capability of amplifying the discrepancy between normal and abnormal network traffic is adopted to detect DDoS attack. The experiments over actual and composite network attack traffic show that the proposed algorithm can detect all the DDoS attacks in both traces. Meantime, the proposed algorithm is capable of detecting DDoS attack quickly and it can be applied in the high backbone network.