Learning control of teleoperation systems that can be utilized in telerehabilition applications is investigated in this thesis. Specifically, considering the fact that in rehabilitation the patient is required to perform a task over and over again, learning controllers are considered as the most feasible solution, in which desired trajectories are periodic with a known period. Since control of teleoperation systems are directly related with the control of robots that are included to the system, learning control of joint space and task space of these robots are simulated in the first part of this study. Joint space learning controller is designed under the restrictions that the robot dynamic model being uncertain and that joint velocities are unmeasurable. Then, a task–space learning controller is designed by considering the fact that the most desired tasks are defined in the end–effector space. Via Lyapunov based stability analysis methods, asymptotic tracking is ensured for both controllers. Numerical simulation results and experimental studies are utilized to illustrate the performance of the designed controllers. In the second part of this thesis, performance of the direct teleoperation and model mediated teleoperation methods under time delays in the communication cahannel are examined in a comparative manner. In direct teleoperation, the information between master and slave systems are exchanged directly, while the model of the environment of the slave system is learnt and integrated at the master side as proxy dynamics in model mediated teleoperation. Experimental studies are realized to evaluate the performance of both of mentioned methods.