Model updating issues with high-dimensional and strong-nonlinear optimization processes are still unsolved by most optimization methods. In this study, a hybrid methodology that combines the Gaussian-white-noise-mutation particle swarm optimization (GMPSO), back-propagation neural network (BPNN) and Latin hypercube sampling (LHS) technique is proposed. In this approach, as a meta-heuristic algorithm with the least modification to the standard PSO, GMPSO simultaneously offers convenient programming and good performance in optimization. The BPNN with LHS establishes the meta-models for FEM to accelerate efficiency during the updating process. A case study of the model updating of an actual bridge with no distribution but bounded parameters was carried out using this methodology with two different objective functions. One considers only the frequencies of the main girder and the other considers both the frequencies and vertical displacements of typical points. The updating results show that the methodology is a sound approach to solve an actual complex bridge structure and offers good agreement in the frequencies and mode shapes of the updated model and test data. Based on the shape comparison of the main girder at the finished state with different objective functions, it is emphasized that both the dynamic and static responses should be taken into consideration during the model updating process.