Stroke will become one of the main burdens of disease and loss of quality of life in the near future. However, we still have not found rehabilitation approaches that can scale up so as to face this challenge. Virtual reality based therapy systems are a great promise for directly addressing this challenge. Here we review different approaches that are based on this technology, their assumptions and clinical impact. We will focus on virtual reality based rehabilitation systems that combine hypotheses on the aftermath of stroke and the neuronal mechanisms of recovery that directly aims at addressing this challenge. In particular we will analyze the, so called, Rehabilitation Gaming System (RGS) that proposes the use of non-invasive multi-modal stimulation to activate intact neuronal systems that provide direct stimulation to motor areas affected by brain lesions. The RGS is designed to engage the patients in task specific training scenarios that adapt to their performance, allowing for an individualized training of graded difficulty and complexity. Although RGS stands for a generic rehabilitative approach it has been specifically tested for the rehabilitation of motor deficits of the upper extremities of stroke patients. In this chapter we review the main foundations and properties of the RGS, and report on the major findings extracted from studies with healthy and stroke subjects. We show that the RGS captures qualitative and quantitative data on motor deficits, and that this is transferred between real and VR tasks. Additionally, we show how the RGS uses the detailed assessment of the kinematics and performance of stroke patients to individualize the treatment. Subsequently, we will discuss how real-time physiology can be used to provide additional measures to assess the task difficulty and subject engagement. Finally, we report on preliminary results of an ongoing longitudinal study on acute stroke patients.