In many applications of wireless sensor networks, location is very important information. Conventional location information comes from manual setting or GPS device. However, manual location setting requires huge cost of human time, and GPS location setting requires expensive device cost and large device size. Furthermore, GPS can not operate in indoor environment. Both approaches are not applicable to localization task of wireless sensor networks. In this paper, an accurate and efficient localization algorithm based on multidimensional scaling (MDS) is proposed in hierarchical network environment. Through localized computation of multidimensional scaling within a cluster, the computation overhead can be distributed to each cluster. On the other hand, MDS-based localization approach requires the estimation of multihop distance. Through restricting estimation of multihop distances within a cluster, the proposed localization algorithm achieves better accuracy and can operate in non-convex network environment. Experimental results reveal that the proposed HMDS localization algorithm outperforms MDSMAP algorithm in terms of accuracy.