Aim: Recent findings suggest that, when reaching or grasping an object, sensory information is encoded in multiple, parallel reference frames. For example, in some proprioceptive assessments, such as the thumb localization task with eyes closed, proprioceptive signals tend to be re-encoded in visual-space. The process of re-encoding sensory signals, from proprioceptive to visual space, is referred to as “cross-reference transformation”. We propose to apply this concept to the study of proprioceptive deficits post-stroke, distinguishing between proprioceptive inputs available for a given assessment, and the potential cross-reference transformations that ensue during task execution. Methods and Results: We reviewed and reinterpreted the outcomes of the most common tests used to assess proprioceptive function in stroke survivors. We identified four main categories of clinical tests, each characterized by a specific processing of proprioceptive signals that may, or may not be re-encoded in different sensory spaces (see figure). This categorization, together with optimal sensory integration modeling, can reconcile the apparently contradictory results of a large number of clinical studies and can explain functionally why different tests lead, in the same patients, to different outcomes. Similarly, reinterpreting the relation between cerebral structures and proprioceptive functions in the light of our theoretical approach provides a novel rationale for an improved stratification of stroke patients. Conclusions: This novel classification of post-stroke sensory deficits may lead to innovative approaches to stroke assessment and rehabilitation. Targeted training of proprioception and/or cross-reference transformations, depending on the patient’s specific of sensory deficit, may be key to improved recovery.