In　this　paper,　the　data-driven　stochastic　subspace　identification　(SSI)　algorithm　is　used　to　identify　the　state　matrix　A　as　a　damage-sensitive　feature.　In　order　to　overcome　the　variety　of　the　state　matrix　A,　a　non-singularity　transposition　matrix　T　is　build.　By　using　matrix　T,　through　which,　matrix　A　is　converted　to　its　observability　standard　matrix.　Then,　matrix　A　is　decomposed　into　m　entries　subsystems　(m　is　the　measure　point　number).　In　order　to　overcome　the　effects　of　computing　model　err,　noise,　environmental　variability　on　the　measured　dynamics　response　of　structures,　the　statistical　pattern　recognition　paradigm　is　introduced　to　the　thesis,　and　the　Mahalanobis　distance　decision　functions　of　the　damage-sensitive　feature　vector　are　adopted,　and　the　β　Novelty　Index　(NI)　is　defined.　The　efficiency　of　the　method　is　verified　using　vibration　measured　data　obtained　from　simulated　simple　beams　and　pre-stressed　concrete　beams　which　were　tested　in　the　laboratory　under　ambient　excitations.