The article reviews how simulation experiments are designed and analyzed using two types of metamodels: low-order polynomial regression and Kriging. The type of metamodel used determines the design of the simulation experiment, which in turn determines the input combinations of the simulation model. For example, first-order polynomial regression should use a "resolution-III" design, while Kriging may use "Latin hypercube sampling". Before applying regression or Kriging, the inputs of a simulation model can be screened through "sequential bifurcation". Optimization of the simulated system can be done using low-order polynomials or Kriging models fitted through sequential designs. "Robust" optimization takes into account uncertainty in some simulation inputs.