Aiming at the limitation that methods based on Kalman filtering framework can deal with the known Gaussian white noise only, an improved method was proposed to deal with the system with heterogeneous noises. The disturbance from an unknown system was classified to Gaussian white noise and unknown but bounded noise which are both added into the state equation and observation equation on the basis of the noise characteristics analyzed. The set-membership filter was employed to improve standard Kalman filtering, and the adjusted value of the gain filtering was obtained by calculating the minimum mean square error of system with two noises. The improved filtering was proposed to incorporate the estimator statistics of the set-membership filter, two noises statistics and the adjusted filter gain. The improved algorithm was applied to an uncertain vehicle navigation system, and the simulation results show that the improved filtering algorithm which overcomes the heterogeneous disturbance noise performs better than the extended Kalman filtering.