Belief propagation (BP) is one of the best-known graphical model for inference in statistical physics, artificial intelligence, computer vision, etc. Furthermore, a recent research in distributed sensor network localization showed us that BP is an efficient way to obtain sensor location as well as appropriate uncertainty. However, BP convergence is not guaranteed in a network with loops. In this paper, we propose localization using generalized belief propagation based on junction tree method (GBP-JT) and nonparametric (particle-based) approximation of this algorithm (NGBP-JT). We illustrate it in a network with loop where BP shows poor performance. In fact, we compared estimated locations with nonparametric belief propagation (NBP) algorithm. According to our simulation results, GBP-JT resolved the problems with loops, but the price for this is unacceptable large computational cost. Therefore, our approximated version of this algorithm, NGBP-JT, reduced significantly this cost, with little effect on accuracy.