A procedure for identifying the dynamic characteristics of a structural system using the multivariate AR (ARV) model is presented in the paper. A least-squares approach to determine the coefficient matrices of the ARV model is proposed, which is a modification of the traditional one. The modification is based on the equivalence between the correlation function matrix for the responses of a linear system subjected to white-noise input and the deterministic free vibration responses of the system. Then, the dynamic characteristics of the system are evaluated directly from the coefficient matrices of the ARV model by adopting the concept behind Ibrahim's system identification technique. The superiority of the proposed procedure over the traditional one is demonstrated through numerical studies on a six-story shear building. Finally, the present identification procedure is applied to identify the dynamic characteristics of a three-span highway bridge from its ambient vibration responses subjected to traffic flow. The validity of the present procedure is confirmed by the excellent agreement between the identified results obtained in the present study and those obtained from an impulse test.