To provide an efficient multiobjective optimizer, an approximation technique based on the moving least squares approximation is integrated into an improved genetic algorithm. In order to use fully, both the a posteriori information gathered from the latest searched nondominated solutions and the a priori knowledge about the search space and individuals, in guiding the search towards more and better Pareto solutions, a gradient direction based perturbation search strategy and a preference function based fitness penalization scheme are proposed. Numerical results are reported to validate the proposed work.