Abstract: The research on stochastic prediction about long-term deflection of reinforced concrete structure is carried in the paper. The response surface approach is adopted to fit the structural responses which cannot be expressed explicitly. Meanwhile, through introducing importance sampling approach into Bayesian inference, the numerical computational dilemma, encountered when the likelihood function has a sharper distribution, is overcome. With prediction of long-term deflection for a reinforced concrete beam, the accuracy of the proposed computational scheme is verified.