Robotic rehabilitation devices have been suggested as a tool to increase the amount of rehabilitation delivered after a neurological injury. Clinical robotic rehabilitation studies of the upper extremity have generally focused on stroke survivors. We present the results of a multi-center pilot study where an upper-limb robotic rehabilitation device (Armeo Spring®, Hocoma AG) was incorporated into the rehabilitation program of 12 subjects with sub-acute cervical spinal cord injury (motor level C4-C6, AIS A-D). Outcomes were measured using two tests of upper extremity function: ARAT and GRASSP. The change in scores for the arm receiving the Armeo training were not statistically significant when compared to the arm not receiving the Armeo training at discharge from therapy and over follow up assessments (8.7 +/- 2.9 compared to 7.4 +/- 2.5 for ARAT at discharge, p = 0.98, and 13.0 +/- 3.2 compared to 13.3 +/- 3.3 for GRASSP at discharge, p = 0.69). Nevertheless, subjects with some minimal (partial) hand function at baseline had a significantly larger increase in GRASSP scores than subjects with no minimal hand function preserved at baseline (19.3 +/- 2.4 compared to 6.6 +/- 4.7, p = 0.02). This suggests that the initial functional capabilities of patients can influence the benefits measured after robotic rehabilitation training and heterogeneous subject populations should be avoided in early phase studies.