In this article, a novel algorithm is proposed for real-time system identification using hierarchical interhealing model classes. One major difficulty in the system identification for large structures is to determine the complexity of the structural model and the number of unknown parameters. One can detect finer damages with more unknown stiffness parameters, but this may cause fluctuating or even unidentifiable results. Although Bayesian model class selection allows one to choose among some prescribed model classes, the number of possible model classes for large structures is huge. In this paper, we propose a new method using hierarchical interhealing model classes. The modeling errors of these model classes can be corrected adaptively according to the data and the results from the optimal model class. This includes not only the errors in the parameters but also the deficiencies of the parametric models. Furthermore, the model classes are established in a hierarchical manner so that the proposed strategy requires only a small number of model classes, yet being able to explore a large solution space. Consequently, the proposed algorithm can handle a large number of damage possibilities while it maintains relatively low computational cost. Two examples are presented, and the results show that the proposed algorithm can detect, locate, and quantify damages reliably and efficiently.