Robust and performing navigation systems for Autonomous Underwater Vehicles (AUVs) play a discriminant role towards the success of complex underwater missions involving one or more AUVs. The quality of the filtering algorithm for the estimation of the AUV navigation state strongly affects the performance of the overall system. In this paper, the authors present a comparison between the Extended Kalman Filter (EKF) approach, classically used in the field of underwater robotics and an Unscented Kalman Filter (UKF). The comparison results to be significant as the two strategies of filtering are based on the same process and sensors models. The UKF-based approach, here adapted to the AUV case, demonstrates to be a good trade-off between estimation accuracy and computational load. UKF has not yet been extensively used in practical underwater applications, even if it turns out to be quite promising. The proposed results rely on the data acquired during a sea mission performed by one of the two Typhoon class vehicles involved in the NATO CommsNet13 experiment (held in September 2013). As ground truth for performance evaluation and comparison, performed offline, position measurements obtained through Ultra-Short BaseLine (USBL) fixes are used. The result analysis leads to identify both the strategies as effective for the purpose of being included in the control loop of an AUV. The UKF approach demonstrates higher performance encouraging its implementation as a more suitable navigation algorithm even if, up to now, it is still not used much in this field.