An improved eigensystem realization algorithm and mode order determination based on singular value decomposition (SVD) is presented. A mode order indicator, called singular value percentage (SVP), is developed according to the weight of selected modes. The indicator is applied to the improved algorithm to determine mode orders in modal identification. The proposed approach includes four main steps: (1) constructing the initial Hankel matrix; (2) eliminating the noise via SVD and obtaining de-noise signal matrix; (3) reconstructing the Hankel matrix by using Cadzow’s algorithm; (4) utilizing the proposed SVP indicator to determine the mode order. Several simulation cases are studied and it is proved that the improved ERA has a better noise resistance ability. The new indicator SVP is effective in mode order determination and false modes elimination. The modal damping ratio can be identified successfully as well. This improved method is also applied to the modal parameter identification of a sedan exhaust system. The effectiveness is proved by comparing identified results with data from LMS system, especially in determining the mode order