Abstract: Metamodel is an effective tool in modeling and optimizing complex systems, it is important to choose appropriate metamodel for specific domain. This paper presents the basic feature and essential background of metamodels, some widely used metamodeling techniques including response surface method, multivariate adaptive regression spline, radical basis function, Kriging and support vector regression are briefly reviewed. Several representative benchmark testing functions are employed to compare the robustness, global and local fitting capacity of these five metamodels. Several influencing factors are investigated, including complexity and nonlinearity of problems, scale and distribution of samples. The effectiveness of metamodel is illustrated in an engineering example.