A new surrogate model based algorithm for global optimization is derived. A metamodel such as a kriging model or a radial basis function model is used to build an interpolant of the objective function. Evaluation points are chosen in such a way that local search and global exploration is balanced. Instead of putting the next point exactly where it is most likely to find the global optimum, the new method also prepares for steps to come by minimizing the total uncertainty of the interpolated function within the most interesting areas of the search space.