It has important theory worth and realistic meaning to perform health monitoring for the in-service civil engineering structure and the infrastructure，in order to discover the damage as early as possible，the feasible preventing and repairing measure will be applied in time. Acceleration and strain response covariance and the Bayesian data fusion technique are used for damage occurrence and damage location identification. From the derivation，the response covariance is the function of modal parameters. Structural damage（stiffness reduction） will lead to the change of the response covariance. When damage detection is based on the change of the response covariance between the damaged and intact structural states without the need of an analytical model，the damage vector will be easily affected by excitation location，measurement noise and so on. In order to improve the results，Bayesian fusion is applied to all damage vectors from different kinds of sensors and many measurements under different environments. A seven-floor frame structure is numerically studied with several damage scenarios，which include single and multiple damages，to demonstrate the applicability and availability of the proposed method. Finally，a simply supported steel beam is tested in the lab to verify the method experimentally. The satisfactory results are obtained. The analysis shows that the proposed method is sensitive to structural damage and robust to noise，and is promising for engineering structures.