Metamodels have been widely used in engineering design to replace the high-fidelity simulations which lead to high computation costs. The accuracy of a metamodel is strongly affected by the sampling strategies. Most strategies use the Kriging as a metamodel since it provides information on the prediction error. In this paper, a new sequential sampling approach is proposed, which is capable of employing many other metamodels: it is not restricted within Kriging. It selects additional points by two ways according to the number of sampling points. The capabilities of the proposed approach have been demonstrated by test problems with various features.