Myoelectric pattern recognition has been traditionally used for the control of robotic prosthetic hands by amputee patients [Scheme, 2011, JRRD]. This technique could also be leveraged as a control paradigm for assistive devices such as neuroprostheses or exoskeletons in people with incomplete paralysis. Recently, pilot demonstrations of gesture recognition based on residual forearm EMG activity have been performed in patients with stroke [Meyers, 2024, JNER] and spinal cord injury (SCI) [Lu, 2019, JNE; Ting, 2021, J Neurophysiol.] using either standard bipolar EMG electrodes or a high-density EMG array. However, both approaches have limitations: standard bipolar electrodes require careful placement, while high-density arrays require complex electronics to process the large amounts of data. A trade-off solution could consist in medium-density EMG systems. Here, we investigated myoelectric gesture recognition in patients with incomplete paralysis after SCI using a novel, easy-to-set-up medium-density EMG system consisting of 64 dry electrodes. We recruited five SCI patients with different degrees of motor impairment, that is MRC grade of finger flexion and wrist extension ranging from 2 to 4. We recorded residual forearm EMG activity using our system while participants attempted different hand gestures, including different grasp types. Offline, we applied decoding algorithms to discriminate between gestures: movements could be classified with an accuracy above 70% regardless of the patients’ motor deficits. These results show that our medium-density EMG system is a promising solution for the control of assistive devices in patients with incomplete paralysis, which could overcome the limitations of existing approaches.