Virtual reality exposure therapy (VRET) is the transform of in vivo exposure therapy by using virtual reality during the treatment. In recent years, VRET has become an attention-grabbing alternative to the traditional treatments of social anxiety. This systematic review gives an overview of the trials regarding the efficacy of VRET in patients with social anxiety. Furthermore, researches on the mechanism during the exposure process are discussed. The application of VRET was divided into three stages. Pilot studies was used to show that VR might be a useful tool for exposure therapy in the treatment of social anxiety. However, these studies are vulnerable to the threats to internal validity such as historical effects and maturational effects for not using controlled group or assigning participants randomly. Several controlled trials was used in the second and generated greater confidence. Although intervention groups received significant improvement compared with the wait-list, all these studies investigated VRET in combination with CBT, therefore the efficacy of VRET as stand-alone treatment is still unknown because the possibility cannot be ruled out that the effects found were caused by CBT rather than VRET. So in the third stage, two studies began to make an effort to single out the effects of pure VRET with exposure components alone. The results indicated that VRET as a stand-alone treatment is valid and can successfully be applied for treatment purpose. Furthermore,the study to evaluate the sustainability of treatment gains also emerge,which enhance the conclusion of the efficacy of VRET. The efficacy of these studies could be explained by the relationship between the authenticity of virtual reality, emotions felt by the users immersed in the virtual environment and the sense of presence. The concept of presence has been identified as the machanism by which virtual reality exposure therapy can treat social anxiety successfully. Different componets of presence was linked to peak fear and the response.There is a positive association between peak fear and realness subscale, while the response during VRE is predicted by involvement subscale. Future research in social anxiety should not only focus on attentional deployment to verbal cues, but also to nonverbal cues such as facial expressions. Moreover,the application of VR is still in its infancy, most social situations targeted public speaking-related anxiety and included only limited verbal interaction (i.e., answering questions). However, the majority of individuals with social anxiety reported more than one fear, emphasizing the need for research on VRET targeting heterogeneous social fears and design more sceneries. Last but not least, we suggest future researches to measure the third perspective of VRET. The application of avatar can affect people`s attitudes, cues and emotions in the social interaction.So researchers could use avatar to explore the factors that might cure people with social anxiety.