Many sensor network applications are based on the knowledge of sensor locations. A priori, some sensors know their position (via GPS). The objective of this paper is to allow positioning of remaining sensors with high accuracy. Existing technologies (e.g. ToA - Time of Arival) allow sensors to calculate ranges with their neighbor nodes. A new and original range-based distributed technique, called AT-Dist, is presented. It uses two important properties: first, it allows to locate sensor nodes while giving them the accuracy of their position. Thus, a sensor node can deduce if its estimated position is close to its real position and contribute to the positioning of others nodes. Second, AT-Dist manages introduction of measure errors which is the main drawback of range-based methods. As a result, AT-Dist locates with an exact or accuracy position many sensors. Simulations show the efficiency of our method in comparison to others.