Action Observation Treatment (AOT) and Virtual Reality (VR) are both effective interventions for upper limb rehabilitation in patients affected by stroke. The present study aims to demonstrate the effectiveness of Action Observation (AO) combined with VR (AO+VR) in upper limb stroke rehabilitation, as compared to a control treatment based on VR alone (CO+VR). A multicentric randomized controlled trial (RCT) with blinded assessment and two investigative arms was conducted. Adult stroke survivors with hemiplegia (N=26) were randomized into AO+VR or CO+VR treatment groups. The treatment consisted in 20 sessions (1 hour each) over 20 working days in a 5-week period. AO+VR group observed daily actions and replicated them in a virtual scenario, while CO+VR group performed the same actions in VR, without imitation. Motor performance was assessed at baseline (T0), after treatment (T1), and at 6 months after treatment (T2) using the Box and Block Test (BBT), Motricity Index and Barthel Index. A subset of N=8 patients also underwent fMRI motor assessment. Both groups showed improvement in BBT scores for both hands. However, in the AO+VR group the improvement in paretic hand use was significantly greater as compared to the CO+VR group at T1 and T2. The other outcome measures showed common improvement in both groups. The fMRI assessments revealed increased activation of the ipsilesional primary motor, ventral premotor and supplementary motor cortex after AO+VR treatment. These findings indicate the effectiveness of AO+VR rehabilitation, suggesting that this intervention could substantially improve the actual VR-based treatments.