The number of people who need rehabilitation increases day by day because of reasons such as laceration, aging, work accidents and etc. Therefore, the need of rehabilitation aids is constantly increasing. There are many research studies about assistive technologies in rehabilitation. Especially, rehabilitation robots have a great importance. Existing rehabilitation robot studies have mostly focused on position and force control. Thus, it is muscular activation that should be evaluated to enhance control results, because the same joint trajectory and/or joint torque can be achieved through different muscular combinations. In this study a muscular activation controlled rehabilitation robot system for lower limbs is proposed. A probabilistic artificial neural network model, which can estimate posteriori probability, was used for discrimination of EMG patterns for robot control with EMG signals.