With the rapid increase in aging population of many developed countries, the need for robotic gait training has grown. This paper presents a novel gait rehabilitation system, where conditions for realizing 'stable gait' are first defined, and then the control system is designed to be adjustable and compliant to patients of different walking abilities, while also satisfying the gait conditions. The proposed controller focuses on the reaction force of the end-effecter connected to the patient's lower legs, and coordinate transformation is used to divide the reaction force into tangential and normal components based on the command trajectory, and a virtual force-railway-like trajectory is constructed to guide the leg movements in the correct direction. A gain η is introduced in the controller to adjust the flexibility of the force-railway. Experiments are done to verify the effectiveness of the proposed controller.