Robotic rehabilitation for individuals with spinal cord injury (SCI) has been shown to be most effective when the user is motivated and mentally engaged in the execution of therapeutic exercises. Consequently, designers of the human-robot interface are challenged with developing control schemes that can detect user intent and maximize their engagement. Electromyography (EMG) is a promising technique to address this challenge. In this thesis, an EMG control scheme for an upper-limb exoskeleton, the MAHI Exo-II, is designed and tested with a population of able-bodied users, as well as SCI affected subjects. The presented scheme utilizes pattern recognition techniques to monitor the user's muscle activation patterns and select their intended direction of motion in single or multiple degree-of-freedom (DoF) movements of the elbow and wrist joints. The results presented demonstrate that the control scheme was simple to use, highly adaptive across a range of subjects, and accurate in directional classification.