To investigate the uncertainty of modal parameters of the long-span cable-stayed bridge，the genetic algorithm is introduced into the traditional fast Bayesian FFT（FBFFT）approach，and the asymptotic approximation with the assumption of high signal-to-noise ratio is used to constrain the search space of parameters of genetic algorithm. Thus，a modal parameter identification approach for the long-span bridge is developed. The recognition efficiency and accuracy of the proposed hybrid method are verified by using the numerical simulation of a cantilever beam. The modal parameters of Sutong Bridge are identified using the proposed method and according to the measured acceleration data. On that basis，the influence of bandwidth factor on the identification accuracy and the uncertainty of modal parameters are investigated. The characteristic of the posterior probability density function （PDF）of modal parameters is analyzed. The results show that every modal parameter of the long-span cable-stayed bridge can be identified effectively using the proposed method. The uncertainty of the identified frequencies and mode shapes is much lower than that of identified damping ratios. By setting the bandwidth factor between 5 and 10，the balance of identification error and uncertainty can be reached. The posterior PDF of identified modal parameters is highly consistent with the Gaussian PDF.