Background. Embedding mirror therapy within a virtual reality (VR) system may have a superior effect on motor remediation for chronic stroke patients. Objective. The objective is to investigate the differences in the effects of using conventional occupational therapy (COT), mirror therapy (MT), and VR-based MT (VR-MT) training on the sensorimotor function of the upper limb in chronic stroke patients. Methods. This was a single-blinded randomized controlled trial. A total of 54 participants, including chronic stroke patients, were randomized into a COT, MT, or VR-MT group. In addition to 20-minute sessions of task-specific training, patients received programs of 30 minutes of VR-MT, 30 minutes of MT, and 30 minutes of COT, respectively, in the VR-MT, MT, and COT groups twice a week for 9 weeks. The Fugl-Meyer motor assessment for the upper extremities (FM-UE; primary outcome), Semmes-Weinstein monofilament, motor activity log, modified Ashworth scale, and the box and block test were recorded at pre-treatment, post-intervention, and 12-week follow-up. Results. Fifty-two participants completed the study. There was no statistically significant group-by-time interaction effects on the FM-UE score (generalized estimating equations, (GEE), P = .075). Meanwhile, there were statistically significant group-by-time interaction effects on the wrist sub-score of the FM-UE (GEE, P = .012) and the result of box and block test (GEE, P = .044). Conclusions. VR-MT seemed to have potential effects on restoring the upper extremity motor function for chronic stroke patients. However, further confirmatory studies are warranted for the rather weak evidence of adding VR to MT on improving primary outcome of this study. Clinical trial registration: NCT03329417.