Abstract: Proposed in this paper is a distributed localization algorithm for wireless sensor network, which takes advantage of the measured distance between neighbors and the coordinates of two-hop nodes. This algorithm is very robust, and it performs well even when the measured distance error is high, which is seldom possessed by localization algorithms under similar conditions. The main idea of this algorithm is to localize the node constrained by many conditions first, so the coordinates calculated are very accurate, after that, some constraints will be incorporated, and so on. It is verified in experiment that this technique can greatly enhance the localization accuracy. This algorithm is composed of three stages. In the first stage, a routing tree is set up, so each node can communicate with root; in the second stage, each node's coordinates are calculated by the order determined by localization priority; in the third stage, all nodes' coordinates are slightly altered to minimize error. The novelty of this localization algorithm explained, in detail, and the reason and effect of each optimization technique are evaluated. The experiments are conducted mainly about the influence of average neighbor number and the distance measurement error on localization accuracy.