High-resolution neuroprosthetic decoding of finger movements and hand gestures could greatly improve the usability of robotic limbs and enable new communication methods such as typewriting. Here, we aimed to understand the motor cortical neural activity during naturalistic hand gestures, to guide the design of hand motion neuroprostheses. Simultaneous neural population activity was recorded from two intracortical Utah arrays (192 channels) in the ‘hand knob’ area of left premotor cortex of a participant with tetraplegia (‘T5’) while he attempted different hand gestures. Trials varied across laterality (left/right hand) and palm position (up/down/sideways). Trials started from a neutral resting position and requested attempting to move (‘move’) to either a natural gesture from sign language or one of 81 different combinations of three movements (flex/extend/idle for four finger groups (ring and little fingers had the same movement)). For the same single finger movement (flexion/extension), changing the hand laterality (left/right) or the hand pose (up/down/sideways) caused neural activity patterns to be largely transformed by a uniform, time-independent shift for the entire duration of the trial. For single finger movements, nearby fingers had similar representation, reflecting the biomechanical constraints of an able-bodied hand. While various gestures with simultaneous finger movements are distinctly represented in the neural activity, we found that the neural representation for individual fingers rotated significantly based on the movement of other fingers. To compensate for these non-linearities, an RNN was trained to decode continuous finger movements, enabling T5 to perform a “center-out” task with independent targets for two finger groups. Overall, insights about the neural activity corresponding to individual and simultaneous movements of multiple effectors would enable neuroprosthetic control of other complex movements that are uniquely human.