Motor imagery combining virtual reality (VR) technique has recently been reported to have an increasingly positive impact on post-stroke rehabilitation. However, there is a common problem that the engagement of patients cannot be confirmed during motor imagery training due to a lack of effective feedback control. This paper proposes a VR-based motor imagery training system for post-stroke rehabilitation, using surface electromyographic (EMG)-based real-time feedback to enable the personalized training and quantitative assessment of participation degree. Three different experiments including assessment experiment, action observation (AO), combined motor imagery and action observation (MI+AO) experiment were performed on 4 post-stroke patients to verify the system. The immersive scenario of the VR system provides a shooting basketball training for bilateral upper limbs. The EMG data of assessment of each participant was collected to calculate the thresholds, which was utilized in the subsequent experiments based on real-time feedback of EMG. The result reveals significant differences of the muscle strength between AO and MI+AO experiments. This demonstrates that the EMG-based feedback is effective to be of use in assessment of participation degree. The primary application shows that VR-assisted motor imagery system has potential to provide personalized and more engaged training for post-stroke rehabilitation.