This paper addresses the problem of beamforming for antenna arrays in the presence of mismatches between the true and nominal steering vectors. A new method for robust beamforming is proposed by minimizing the array output power while controlling the array mainlobe response. Due to the presence of the non-convex response constraints, a new approach based on iteratively linearizing the non-convex constraints is proposed to reformulate the non-convex problem to a series of second-order cone programming (SOCP) subproblems, each of which can be optimally solved by well-established convex optimization techniques. Simulation results show that the proposed method offers better performance than conventional methods tested.