This paper presents a structural identification procedure for the finite element (FE) model updating and the joint damage assessment in steel frame structures. It consists of four parts: i.e., dynamic and static tests, modal identification using the stochastic subspace identification method, baseline FE model updating using the inverse perturbation technique, and joint damage assessment using the neural networks technique. The Tikhonov regularization algorithm is employed to reduce the ill-posedness during the FE model updating. The substructural identification technique using limited measurement data is presented for large and complex structures. The effectiveness of the proposed procedure has been verified through an experimental study on a steel frame structure with 2-days and 4-stories. It has been found that the proposed method is efficient for the FE model updating and the joint damage assessment.