In order to enhance the efficiency of stochastic vibration analysis for train-bridge coupling systems, this paper proposes a novel approach based on the parallel adaptive enhanced (PAE)-surrogate model. First, an initial surrogate model is established to predict the extreme values of dynamic responses in the train-bridge coupling system using a small number of training samples. Second, a multipoint adaptive sampling method is employed to determine a set of new samples that provide more information. The theoretical extreme values of the dynamic responses corresponding to the new samples are cal...