Constant interactions of motor and sensory signals occur at many levels in the nervous system to support animals’ adaptive behaviors. During movements, motor brain regions not only send motor commands to descending spinal pathways, but also corollary discharge (CD) signals to other brain areas to modulate sensory processing. Theoretical work suggests the sensory predictions conveyed by CD signals are combined with sensory feedback to form an optimal estimate of the body state. However, neural underpinnings of this integration in cortex remain elusive. Here, we used a neural decoding-based approach to characterize the interaction of CD and sensory feedback signals in area 2 of somatosensory cortex, a brain area that processes proprioceptive information. One challenge in understanding the interaction of CD and feedback signals is that they are temporally overlapping and thus neurons have mixed representations during movements. We here disentangled the two signals within population activity via their different onset times relative to movement, and then validated the signal identities during passive perturbations. In doing so, we discovered the two signals occupy orthogonal neural subspaces, enabling their simultaneous yet separate representations at the neural population level. To then explore how CD signals influence sensory processing, we asked how body state estimates change when using CD or feedback signals alone versus in combination. Consistent with theoretical propositions, while the two signals can separately decode movement kinematics at distinct time shifts relative to movement, their integration allows for accurate kinematic estimation before feedback information becomes available. Our simulations recapitulated this finding by showing that signal integration helps overcome system noise. Overall, our results show a neural population-level encoding strategy of corollary discharge and sensory feedback signals, provide evidence for their integration in somatosensory cortex, and demonstrate a broadly applicable approach to dissect motor and sensory signals within neural population data.