Determining where a given sensor is physically located is a challenging issue. In this paper, we address the localization problem where, initially, a certain number of sensors are aware of their positions (either from GPS or by being hand-placed). These nodes are referred to as anchors. Our goal is to localize all sensors with high accuracy, while using a limited number of anchors. So, we focus on localization techniques based on angle of arrival information between neighbor nodes. This paper proposes an original angle-based localization technique, called AT-Angle, which allows to verify two important properties: first, a sensor node can eliminate wrong received information about its position; second, it deduces if its estimated position is closed to its real position. In this last case, the sensor node becomes an estimated anchor and contributes to the positioning of others nodes. By varying the density and the error rate, simulations show that AT-Angle achieves good precision for located nodes despite the introduction of errors on positions and the small number of anchors.