An improved nonlinear filter is proposed in the framework of the cubature Kalman filter (CKF) with uncompensated biases. This filter can be applied for the nonlinear system with unknown random biases, which are unable to be modeled in practical situations. The proposed method can decrease the state estimation error and demonstrate the excellent numerical stability in the process of the filtering. The performance is verified by Matlab simulations in the context of the radar tracking.