This paper investigates wireless sensor networks where a small percentage of nodes are assumed to know their location a priori. These reference nodes enable absolute localization of other nodes in direct neighborhood. Having estimated their location, these nodes in turn provide their location to other nodes within transmission range. Therefore, location information spreads throughout the network. Consequently, in later state of the network, unknowns desiring to determine their location, or to improve it, will be able to choose from a large pool of nodes with known or estimated locations, which we refer to as beacons. We investigate a method to select a subset of beacons to minimize the error of localization. Regarding Cramer-Rao-Lower-Bound on localization error, the method proposed constitutes a significant improvement in comparison with the often used nearest-neighbors approach.