This paper describes a development of a method for motion recognition using frequency distribution considering motion transition in order to build an exoskeleton robot system for assistance in rehabilitation. Firstly, a method of recognition based on calculation of differences in frequency distribution between template data and current one has been proposed. Additionally, two sets of parameters were assembled to deal with keeping stiffness in sustained action and compliance in motion transition. Furthermore, method for detection of motion transition has been proposed and implemented into the system. As the result, the recognition ratio has been successfully increased to eighty percent by using the above-mentioned method.