In this paper, we propose a system for training of stroke patients with unilateral neglect by using technology of virtual reality (VR). The proposed system is designed to compensate for unilateral neglect. This system contains the calibration of unilateral neglect and the training of this disease. The calibration procedure is implemented by aligning the virtual object at a subjective middle line. The training procedure is implemented by completing the missions that are used to keep the virtual avatar safe during crossing the street in a virtual environment. The results of this study show that the proposed system is effective to train unilateral neglect. The left to right ratio scores extracted from this system gradually decrease as the sessions of training are repeated. To validate the VR system parameters, the parameters are analyzed by correlation with those of traditional unilateral neglect assessment methods (such as the line bisection test and the cancellation test).