This contribution presents a recursive algorithm for state estimation of nonlinear systems using ellipsoidal bounds on the process and observation noises. A novel approach based on state bounding techniques and on the classical Extended Kalman Filter with switching gain is proposed. A particular parameterization of the algorithm is introduced to increase performances and to characterize the set of state estimates compatible with the noises bounds. Simulation results on a £fth-order two-phase nonlinear model of an induction motor are also given.