The recorded motion from (pathological) tremor patient may consist of the involuntary tremulous component and the intended motion. These two components have to be separated so that the actuation part will be able to suppress only the tremor. This paper proposes an algorithm to remove the intended motion by using an extended Kalman filter with the help of adaptive high-pass filter. The effectiveness of the algorithm is also shown in the presence of stimulation artifacts. It is part of the active pathological tremor compensation project for human upper limb.