This paper describes an application of Kalman filtering to the measurement of solid friction. In the previous report, we proposed an extended Kalman filter algorithm for the estimation of the friction based on the model where the friction is an unknown exponential function of the velocity. simulation studies have shown that the friction can be estimated satisfactorily if the model is exact. In the present paper, we consider the case where the model used in the algorithm does not represent the true friction characteristics. It is shown by extensive simulation studies that the proposed algorithm is still effective unless the true friction characteristics is far from the assumed model and that the performance of the proposed algorithm is superior to that of the Kalman filter algorithm without a model for the friction characteristics.