Modal　parameter　identification　of　ambient　vibration　system　need　extract　information　from　data　using　signal　process　method.　In　this　paper,　a　newly　developed　signal　processing　technique,　Empirical　Mode　Decomposition　(EMD),　is　implemented　to　decompose　data　set　into　several　intrinsic　mode　functions　(IMF)　by　a　procedure　called　sifting　process.　By　using　intermittency　frequency,　the　sifting　process　can　overcome　the　difficulty　of　mode　mixing　and　decompose　data　to　modal　response　functions.　Then　the　stochastic　subspace　identification　(SSI)　method　is　used　to　identify　the　modal　parameters　of　civil　engineering　structures.　The　process　combines　the　advantages　of　those　two　methods,　it　can　process　nonlinear　and　non-stationary　signal　generated　by　ambient　vibration　measurements.　And　the　structure　modal　parameters　are　easily　identified　ignoring　the　influence　of　other　modal　frequency　components　and　the　influence　of　fake　frequencies　and　unwanted　noise　by　proposed　EMD-based　stochastic　subspace　identification.　A　case　study　Beichuan　arch　bridge　using　real　operational　acceleration　signals　excited　by　ambient　vibration　is　presented　to　show　the　applicability　of　the　present　technique.