In this paper we discuss the performance of different metamodeling techniques in Type II robust design. Five space-filling experimental designs and five correlation functions are compared in their capability for developing accurate kriging models. We observe that in our case study (a vehicle body), the orthogonal array is the best experimental design and the cubic correlation function is the best correlation function. Then the kriging models and response surface models for system response are determined and the solution obtained using compromise Decision Support Problem. We note that: 1) the goal formulation method of minimizing variance at the center point may be inappropriate in some situation, and 2) more robust solutions can be obtained with kriging models than with response surface models.