This work presents and compares several cooperative navigation solutions for formations of autonomous underwater vehicles, equipped with depth sensors and capable of taking bearing measurements to their neighbors under a certain measurement topology. Two approaches based on the extended Kalman filter are described, one centralized and the other decentralized, which has the advantage of requiring much less communication and computational complexity with minimal degradation of the produced estimates. Additionally, four other Kalman filter implementations, based on systems with linear dynamics using artificial measurements, are also described, one centralized and the remaining ones decentralized. The performance of these algorithms, under both acyclical and cyclical measurement topologies, is compared using Monte Carlo simulations, whereby both the mean error and root-mean-squared-error (RMSE) of the computed navigation estimates are presented.