Buffeting of long-span bridges caused by the wind turbulence could result in problems of large deformation, fatigue, traffic safety and user comfort. The calculation of buffeting responses is greatly affected by multiple uncertainties, especially the randomness of flutter derivatives and structural damping. In buffeting analysis, these uncertainties are typically propagated using the brute-force Monte Carlo (MC) method, which requires enormous computational resources for a complicated structure involving multiple uncertainties. This study develops an efficient framework based on surrogate models to account for these uncertainties in buffeting responses and the assessment of structural fragility in a mixed climate. Two surrogate models, Kriging and polynomial chaos expansions (PCE), are applied in this framework. Comparison with the direct MC method shows that the Kriging model rather than the PCE model is the proper surrogate model, and the surrogate model contributes significantly to saving computing time from 17 h to 1 min for MC simulations.It is also observed that uncertainties propagated from structural parameters to responses will be more notable as the wind speed increase. Buffeting fragility curves of this bridge show that it’s easier for responses in acceleration to achieve and exceed thresholds, indicating that performance related to user comfort might not be satisfied. By introducing the probability distributions of non-typhoon and typhoon winds at the site of the bridge, it is found that considering single climate may underestimate structural risk. The framework based on surrogate models in this paper can be further generalized to additional PBWE frameworks addressing different wind and structural engineering issues.