In wireless sensor networks, there is often a need for nodes to find their position. This process is referred to as localization. Many methods have been proposed for this purpose, but they typically suffer from one of two major problems: Either, they are inaccurate for noisy measurement data, or they require a considerable amount of computation. In this paper, we present a method based on recursive estimation of position from angle-of-arrival measurements by directional antennas. The method computes a new position estimate for every new measure ment, using a Kalman filter. Computation is fast and is performed entirely locally. No complex datastructure needs to be maintained. A prominent feature of the proposed method is that it applies only a linear Kalman filter.