Particle filters have become an increasingly useful tool for recursive Bayesian state estimation, especially for nonlinear and non-Gaussian problems. Despite the large number of papers published on particle filters in recent years, one issue that has not been addressed to any significant degree is the robustness. This paper presents a deterministic approach that has emerged in the area of robust filtering, and incorporates it into particle filtering framework. In particular, an ellipsoidal set membership approach is used to define a feasible set for particle sampling that contains the true state of the system, and makes the particle filter robust against unknown but bounded uncertainties. Simulation results show that the proposed algorithm is more robust than the regular particle filter and its variants such as the extended Kalman particle filter.