Background Many contemporary systems for neurorehabilitation utilize 3D virtual environments (VEs) that allow for training patients‘ hand or arm movements. In the current paper we comparatively test the effectiveness of two characteristics of VEs in rehabilitation training when utilizing a 3D haptic interaction device: Stereo Visualization (monoscopic vs stereoscopic image presentation) and Graphic Environment (2.5D vs 3D). Method An experimental study was conducted using a factorial within-subjects design. Patients (10 MS, 8 CVA) completed three tasks, each including a specific arm-movement along one of three directional axes (left-right, up-down and forward-backward). Results The use of stereoscopy within a virtual training environment for neurorehabilitation of CVA and MS patients is most beneficial when the task itself requires movement in depth. Further to this, the 2.5D environment yields the highest efficiency and accuracy in terms of patients‘ movements. These findings were, however, dependent on participants‘ stereoscopic ability. Conclusion Despite the performance benefits of stereoscopy, our findings illustrate the non-triviality of choices of using stereoscopy, and the type of graphic environment implemented. These choices should be made with the task and target group, and even the individual patient in mind.