Severe paresis of the contralesional upper extremity is one of the most common and debilitating post-stroke impairments. The need for cost-effective high-intensity training is driving the development of new technologies, which can complement and extent conventional therapies. Apart from established methods using electromechanical devices, immersive virtual reality (iVR) systems hold promise to provide cost-efficient high-intensity arm training. We investigated whether iVR-based arm training yields at least equivalent effects on upper extremity function as compared to an electromechanically assisted training in stroke patients with severe arm paresis. 52 stroke patients with severe arm paresis received a total of ten daily group therapy sessions over a period of three weeks, which consisted of 20 min of conventional therapy and 20 min of either electromechanically assisted (ARMEOSpring®) or iVR-based (CUREO®) arm training. Changes in upper extremity function was assessed using the Action Research Arm Test (ARAT) and user acceptance was measured with the User Experience Questionnaire (UEQ). iVR-based training was not inferior to electromechanically assisted training. We found that 84% of patients treated with iVR and 50% of patients treated with electromechanically assisted arm training showed a clinically relevant improvement of upper extremity function. This difference could neither be attributed to differences between the groups regarding age, gender, duration after stroke, affected body side or ARAT scores at baseline, nor to differences in the total amount of therapy provided. The present study results show that iVR-based arm training seems to be a promising addition to conventional therapy. Potential mechanisms by which iVR unfolds its effects are discussed.