Analysis of the response statistics of randomly mistuned bladed disk usually is expensive due to the Monte Carlo simulations.The polynomial response surface method and Kriging metamodel are used to predict the amplification factor of the mistuned system in this paper.To improve the accuracy of the surrogate model,the expected improvement (EI) Kriging is proposed to handle the complex nonlinear relations between the random mistuning and the amplification factor.A lumped parameter bladed disk is used as example to show the effectiveness of the surrogate models.The results indicate that,Kriging model with EI criterion performs much better than polynomial response surface method in accuracy.