This paper is concerned with robust strategies to perform a model-based damage detection of multistory buildings effectively. Theoretically, the model-based damage detection can be reasonably carried out through vibration monitoring and system identification. The change in modal properties due to damage is identified accurately by time series model if structural vibrations are measured appropriately. Then, the reduction of local stiffness is estimated correctly from the change in modal properties by using sensitivity equations. However, in practice, it is not an easy task to pinpoint the location and extent of local damage because of noise contamination, system nonlinearity, estimation error, model uncertainty and so on. To overcome these problems, several robust strategies are presented in the model-based damage detection of multistory buildings. The basic idea consists of decreasing the number of physical parameters to be estimated, increasing the number of modal properties with good accuracy, enhancing sensitivity to damage and reducing noise and nonlinear effects. The effectiveness and limitations of these strategies are discussed through a series of shaking table tests of two small-scaled test structures.