An accurate structural model is crucial for many research areas such as structural response prediction, damage identification, and structural health monitoring. However, structural models built based on pure modeling assumptions without including experimental information cannot well reflect actual structural dynamic characteristics. It is necessary to identify models using experimental data. Moreover, uncertainties arising from measurement noise and modeling assumptions make structural identification in practice ill-conditioned, and they need to be addressed. Therefore, developing an efficient method that can identify accurate structural models while quantifies the identification uncertainties has always been an important topic in the field of civil engineering. Bayesian structural identification is one of the most promising and widely recognized approaches in this field. However, current Bayesian structural identification methods suffer from susceptibility to local optima and low efficiency in searching the parameter space. To address these limitations, this paper proposes a Bayesian structural identification method using subset simulation. Based on the Bayesian probabilistic framework, structural identification is formulated as a problem of calculating the probability of a rare event. The extremely small probability of a rare event is then transformed to a series of more frequent intermediate events with larger probabilities. This transformation makes the efficient search of high-dimensional parameter space possible, and it also ensures locating possible equivalent most probable values of parameters. The proposed method is applied for structural identification of a frame structure. The results demonstrate that the identified structural model can accurately predict the actual structural dynamic properties.