Structural health monitoring(SHM)has become a hot and intensively researched field in civil engineering.Thereinto,damage identification play an important role in maintaining structural integrity and safety.Many effective methods have been proposed for damage identification.However,accurate global identification of large real-world structures is not easy due to their complex and often unknown boundary conditions,nonlinear components,insensitivity of global response to localized damages,etc.Furthermore,global identification often requires lots of sensors and involves large number of unknowns.This is costly,rarely feasible in practice,and usually yields severely ill-conditioned identification problems.Substructuring approach is a possible solution:substructuring methods can focus on local small substructures;they need only a few sensors placed on the substructure and yield smaller and numerically much more feasible identification problems.This paper proposed an improved substructure method using local free response for substructure damage identification.The virtual supports are constructed by Substructure Isolation Method(SIM)using the linear combination of the substructural responses.The influence of the global errors is isolated by adding the virtual supports on the main degree of freedoms(DOFs)of the substructure.Through the correlation analysis,the substructural modes are selected and used for damage identification of the substructure.A plain model of cable stayed bridge is used for the verification of the proposed method. Structural health monitoring (SHM) has become a hot and intensively researched field in civil engineering. Thereinto, damage identification play an important role in maintaining structural integrity and safety. Many effective methods have been proposed for damage identification. Hosted, accurate global identification of large real-world structures is not easy due to their complex and often unknown boundary conditions, nonlinear components, insensitivity of global response to localized damages, etc. More than that of global identification often requires lots of sensors and involves large numbers of unknowns.This is costly, rarely feasible in practice, and usually productive severely ill-conditioned identification problems. Substructuring methods can focus on local small substructures; they need only a few sensors placed on the substructure and yield smaller and numerically much more feasible identification problems.This paper proposed an improved substructure metho d using local free response for substructure damage identification. The virtual supports are constructed by Substructure Isolation Method (SIM) using the linear combination of the substructural responses. The influence of the global errors is isolated by adding the virtual supports on the main degree of freedoms (DOFs) of the substructure. Through the correlation analysis, the substructural modes are selected and used for damage identification of the substructure. A plain model of cable stayed bridge is used for the verification of the proposed method.