SUMMARYStructural identification based on analysis of vibration signals is usually formulated as an optimization problem. Nevertheless, identifying a structural system as a whole often involves a great number of unknowns and degrees of freedom, resulting in considerable convergence difficulty and expensive evaluation of objective function. To handle these issues, this paper aims to propose a frequency domain substructural identification method applicable to arbitrary excitations. This is done by introducing the exponential window method in the formulation so that the effect caused by initial condition is alleviated without using zero padding. The proposed identification method is verified through both numerical and experimental studies. It is shown that significant reduction in computer time can be achieved in addition to remarkable identification accuracy. The experimental study also indicates that the proposed strategy is effective in identifying small structural changes. Copyright © 2011 John Wiley & Sons, Ltd.