DDoS　detection　has　been　the　research　focus　in　the　field　of　information　security.　Existing　detecting　methods　such　as　Hurst　parameter　method　and　Markov　model　must　ensure　that　the　network　traffic　signal　f　(t)　is　a　stationary　signal.　But　its　stability　is　just　a　regular　assumption　and　has　no　strict　mathematical　proof.　Therefore　methods　mentioned　above　lack　of　reliable　theoretical　support.　This　article　introduces　Hilbert-HuangTtransformation(HHT).　HHT　does　not　need　to　be　based　on　signal　stability,　but　it　monitors　the　similarity　between　Hilbert　marginal　spectrums　of　adjacent　observation　sequences　so　as　to　realize　DDoS　detection.　The　method　is　experimented　on　DARPA　1999　data　and　simulating　data　respectively.　Experimental　results　show　that　the　method　behaves　better　than　existing　Hurst　parameter　method　in　distinguishing　both　the　normal　and　the　attacked　traffic.