In order to resolve the shortcomings of the traditional Kalman filtering, a new method is presented in which the fuzzy reasoning system is combined with the traditional Kalman technology. By monitoring the covariance between abstract residual and actual residual, this algorithm modifies recursively the gain coefficient of the filter so as to adaptively control the Kalman filter. Finally the optimal estimate is achieved. The computer simulation results of the INS / GPS integrated navigation system indicate that the algorithm is effective and practical.