A parallel model identification scheme is presented to detect member damage of multistory buildings. Three different mathematical models are used in the identification process: the frame model with member-end rotational springs for detecting member damage, the frame model with joint rotational springs for detecting joint damage and the flexural-shear lumped mass model for detecting story damage. The natural frequencies and mode shapes of a target building are identified from acceleration records measured at each story by ARX model. The changes in modal properties before and after damage events are used to define objective functions for each mathematical model. The objective functions are minimized to estimate the location and extent of story, joint and member damage, respectively, by simple genetic algorithm. The proposed scheme is applied to acceleration records of shaking table tests for different beam damage conditions. Based on the identification results, the usefulness and limitations of the proposed scheme are discussed.