This paper proposes a recursive method to determine the process and measurement noise covariance matrix in the extended Kalman filter in application of bearing-only target tracking. One of the requirements of Kalman filters is knowledge of process and measurement noise covariance matrices. If the inappropriate choice of covariance, the filter performance is affected and even there is the possibility of divergence. In this paper, a recursive structure to adapting noise covariance is presented that unlike the conventional methods, instead of direct adapting covariance matrices, based on steepest descent adapting rule structure parameters are adapted. This increases the reliability of the adaptive method and non-negative condition of some of covariance matrix elements to be resolved. To evaluate the performance of proposed method, the bearing-only target tracking scenario is considered. To compare the proposed approach, three adaptive covariance common methods is used that simulation results show that the reliability and efficiency of the proposed method.