This paper addresses the determination of the 18 flutter derivatives of bridge decks from three degrees-of-freedom (3DOF) free vibration data using an improved stochastic search algorithm (ISSA) combined with the unified least-squares (ULS) method.The ISSA is capable of circumventing the local optimum dilemma in pursuing the optimal solution experienced in the traditional ULS method.The validity and accuracy of the ISSA are demonstrated by one numerical example and two long-span, cable-stayed, bridge deck sections.The attractive merit of using different lengths of vertical, torsional, and lateral vibration data in flutter derivatives identification is investigated.The identification error and modal participations in flutter are easily examined through a decomposition of modal components from the original vibration data.The underlying complexities in aeroelastic parameter identification are studied, and the causes of low accuracy of some flutter derivatives are unveiled.Based on the comparative investigation on the aerodynamic characteristics of typical streamlined and bluff bridge decks, an improved understanding of the coupled bridge flutter is achieved.