Metamodeling techniques allow the construction of surrogate design models for the purpose of design exploration such as variable screening, optimization and reliability. The metamodel itself can be used instead of the numerical or analytical physical model as a fast surrogate model. Before the metamodel is available, an initial data set, which is often generated with a Design of Experiment (DoE) scheme, has to be generated with the real physical model with a defined input parameter definition and a suitable design scheme. Based on the evaluated responses of the physical model, we distinguish between scalar, signal and field outputs. Once the input and responses data are available, different metmodel types can be tested and assessed according to the approximation quality. A general framework for an automatic model testing and selection is the Metamodel of Optimal Prognosis in optiSLang. In the following chapter, first the metamodels for scalar outputs are discussed. Later we present different measures and procedures for analyzing the model quality, sensitivity evaluation and adaptation strategies.