In this paper, a new frequency-weighted model reduction method with an a priori error bound is proposed. The method is a generalization of Enns technique and yields stable models even when both input and output weightings are included. The proposed method is compared with other existing methods using numerical examples.