Disturbance observer is significantly becoming the preferred approach since it offers distinct advantages of improving robustness of the force control and accuracy of the force estimation. Current and velocity information are the reference inputs of the disturbance observer whereas the external force estimation is the output. Generally, velocity information without noise effects can be calculated by using low-pass filter but the bandwidth of the data is limited. To obtain wide bandwidth of force sensing with low noise, we have utilized the Kalman filter technique to estimate the motor velocity with the measured data from both linear encoders and acceleration sensors. Through the combination of the linear encoder with an acceleration sensor measurement, a velocity reference containing improved bandwidth can be achieved. This paper will demonstrate that Kalman-filter-based disturbance observer as applied to bilateral control system can provide wide bandwidth superior to the conventional one. In order to ensure the improvements of the proposed method, some experimental results of bilateral control is confirmed theoretically.