OBJECTIVES: To investigate the effect of virtual reality (VR) rehabilitation on upper extremity motor performance of patients with early stroke. DESIGN: Pilot randomized controlled trial. SETTING: Rehabilitation wards. PARTICIPANTS: Twenty three adults with stroke (mean age (SD) = 58.35 (13.45) years and mean time since stroke (SD) = 16.30 (7.44) days). INTERVENTIONS: Participants were randomly assigned to VR group (n=11) or control group (n=12). VR group received nine 30 minutes upper extremity VR therapy in standing (five weekdays in two weeks) plus conventional therapy, which included physical and occupational therapy. Control group received only conventional therapy, which was comparable to total training time received by VR group (mean training hours (SD):VR = 17.07 (2.86); control = 15.50 (2.79)). MAIN OUTCOME MEASURES: The main outcome measure was the Fugl-Meyer Assessment (FMA). Secondary outcomes included Action Research Arm Test, Motor Activity Log and Functional Independence Measure. Results were taken at baseline, post intervention and 1-month post intervention. Participants' feedback and adverse effects were recorded. RESULTS: All participants improved in FMA scores (mean change (SD) = 11.65 (8.56), P<.001). These effects were sustained at one month after intervention (mean (SD) change from baseline = 18.67 (13.26), P<.001). All other outcome measures showed similar patterns. There were no significant differences in improvement between both groups. Majority of the participants found VR training useful and enjoyable, with no serious adverse effects reported. CONCLUSION: Although additional VR training was not superior to conventional therapy alone, this study demonstrates the feasibility of VR training in early stroke.