A new system noise covariance modification algorithm is proposed in order to avoid the problem of degraded performance of the filter due to the incorrect statistics of the system noise. Combined with the Extended Kalman Filter (EFK)、Unscented Kalman Filter (UKF) and Divided Difference Filter (DDF), adaptive nonlinear Kalman filters are developed. The algorithm is applied in nonlinear measurement electro-optical tracking system and the performances of the adaptive nonlinear Kalman filter is compared with the basic nonlinear Kalman filters. The Matlab simulation results show that the filter can modify system noise covariance in real time, efficiently avoid the above problem and the performance outperforms the basic nonlinear Kalman filters.