Here we investigate whether torque or myoelectric control was more efficient during active, robotic-assisted therapy. Eleven hemiparetic stroke patients employed the RehaArm-robot, which consist of an exoskeleton that supports the entire arm and measures in real time the joint position and moment of the shoulder, in four sessions of one hour on consecutive days. At each session, the patients repeatedly performed basic movements of the shoulder in passive and active mode. During the passive mode, subjects learnt the movements. During the active mode, subjects were asked to complete 40 task repetitions in 20 min for each modality, torque and myoelectric control. The number of repetitions achieved, completion rate (CR), was tracked for each control modality as well as subjects' opinion about the ease of use of each modality. The results showed that the severe-to-moderate group (Fugl-Meyer Motor Assessment of the Upper Extremity<=40) achieved a significantly higher CR in myoelectric control than in torque control (p<0.05). For the mild group (>40), the CR was very similar for both control modalities. Subjectively, the two groups considered both control modalities similarly easy to use, being the myoelectric control slightly easier (higher median and middle fifty values). These results support the higher efficacy of the myoelectric control for active, robotic-assisted therapy.