The tight integration of Global Navigation Satellite Systems (GNSSs) and low-cost Ultra-Wide Band (UWB) is a prospective positioning solution for autonomous mobile robots that operate in harsh environments with poor satellite visibility. Thanks to the complementarity of the two systems in terms of coverage and ranging performance, the UWB nodes can be used as anchors providing additional ranging measurements. However, the selection of the integration scheme may be a critical issue since high-accuracy positioning performance has to be traded off with the computational complexity of the implementation. This paper compares the performance of two common Bayesian filtering algorithms - the Extended Kalman Filter (EKF) and the Sequential Importance Resampling Particle Filter (SIR-PF) - for the GNSS/UWB tight integration in a dynamic environment. Considering the error sources triggered by the linear approximation employed in the EKF, simulation results show that the performance of the EKF deteriorates more than the SIR-PF when the user's kinematics changes rapidly and when the user gets close to the UWB anchor. Compared to the EKF, the SIR-PF can therefore guarantee superior positioning accuracy even if at the cost of higher computational complexity.