Few rehabilitation methods have proven their efficacy in increasing sensori-motor recovery and/or function of the upper limb (UL) after stroke. Video games (VGs) are promising tools in this indication. OBJECTIVE: To compare UL rehabilitation by using VGs and conventional rehabilitation (CR) in patients with sub-acute stroke. DESIGN: Single-blind, multicentric trial, with central randomization and stratification by center. SETTING: Physical and rehabilitation medicine departments of 2 university hospitals. PARTICIPANTS: Adults within 3 months after a first vascular cerebral accident, with UL Fugl Meyer Score (UL-FMS)<30/66 and without major cognitive impairment. INTERVENTION: A 45-min additional session of conventional occupational therapy (OT) or a VG-based OT session as add-on therapy to usual rehabilitation programs, 5 days/week for 6 weeks. MAIN OUTCOME MEASURES: Primary outcome: UL-FMS. Secondary outcome: Box and Block Test (BBT), Wolf Motor Function test (WMFT), Motor Activity Log (MAL), Barthel Index and quality of life (SF-36). RESULTS: We included 51 patients (20 women) at a mean (SD) of 27.2 (19.4) days post-stroke (mean age 58 years [range 24-83]), 26 in the CR group and 25 in the VG group (23 in each group at 6-month follow-up). The mean duration of the additional rehabilitation session was similar in both groups: 29.3 (4.3) vs 28.0 (4.4) min in CR and VG groups. Shoulder pain occurred in 4 patients in the VG group versus 7 in the CR group. At day 45, gain in UL-FMS did not significantly differ between the groups (CR mean 17.8 [14.6] vs VG 24.1 [14.8]; P=0.10), whereas gain in BBT was doubled in the VG group (CR 7.4 [12.2] vs VG 15.7 [16.3]; P=0.02). At 6-month follow-up, the study was inconclusive about between-group differences in UL-FMS, BBT and other criteria. Post-hoc analysis showed that gains in UL-FMS or BBT were significantly higher in the VG than CR group for patients included within 30 days post-stroke. CONCLUSION: In general, we cannot conclude that video gaming and conventional OT led to different long-term sensorimotor recovery of the UL after sub-acute stroke. However, when applied within the first month after stroke, video gaming was more efficient than conventional rehabilitation on both sensorimotor recovery and gross grasping function.