Most of the existing virtual reality (VR) Cardiopulmonary Resuscitation (CPR) self-instruction training methods are based on low embodied and noninteractive 360-degree videos, and the effects of the highly immersive embodied VR CPR training system lack rigorous experimental verification. In our study, we compare the learning outcomes of self-instruction CPR training based on a highly immersive and interactive VR system (experimental group) with selfinstruction training based on 2D videos (control group) in terms of willingness to perform CPR, knowledge of CPR, self-efficacy and CPR test performances. We conducted a prepost betweengroup experiment in a local college in Yixin, Jiangsu Province, China. Sixty undergraduate students (30 male and 30 female) aged from 18 to 25 were randomly assigned to the video group or the VR group. There were no significant differences in demographic variables and baseline pretest between the two groups. Both groups received training on cardiac compressions and automated external defibrillator (AED) assisted CPR. Our results revealed that both the VR and video methods significantly improved the participants’ CPR knowledge, self-efficacy and willingness to perform CPR. Our results showed that the immersive VR group had a significant disadvantage in their grasp of compression depth compared with the non-immersive media group. The correct compression depth of the video group was significantly higher than the VR group, and there was no difference between the two groups in terms of compression speed and full rebound rate. Our study suggested that both self-instruction training based on VR and self-instruction training based on video were effective methods, and the highly immersive and embodied VR method did not achieve better results than the video method.