We have shown that motor intentions can be decoded from human posterior parietal cortex (PPC) to enable a brain-machine interface. However, we have yet to determine the coordinate frames in which these intentions are encoded. Are intentions coded with respect to the subject’s gaze, current imagined limb position, intended target location, or a combination of these. Or, is it encoded in world or body coordinates in which case gaze and imagined limb position have no effect. Knowing the answers to these questions has the potential to improve implementations of closed-loop neural prosthetic applications by ensuring that neural decoding algorithms are tailored to the way intention related information is actually encoded in the neural population. Here we explored the reference frames of imagined motor intentions in neural populations recorded in parietal cortex (AIP and SMG), primary somatosensory cortex (S1), and an area near ventral premotor cortex (PMv) of two tetraplegic subjects participating in a human clinical study.