Abstract: The determination of correlation parameters is the key point for constructing Kriging model. It is discussed that the optimum result will depend on the starting points to search if the correlation parameters are determined using conventional numerical optimization methods, such as pattern search method. Then the global optimums of correlation model parameters are obtained by Genetic Algorithms (GA). The problem of Kriging construction depending on the starting points is solved. Additionally, using GA with the improved Kriging model the system can be globally optimized based on the approximate model of the system.