In the past two decades, bilateral teleoperation with haptic feedback has attracted great research and application interests in both robotics and other areas. Initially triggered by the need to handle dangerous and remote distance tasks such as nuclear materials manipulation and space exploration, bilateral teleoperation has found its way into other applications as a result of development of control theory, robotic technology (both hardware and software) and latest breakthrough in artificial intelligence and machine learning. Consequently, bilateral teleoperation is found facing new challenges brought by these new applications. One major and obvious change is the working environment for the slave manipulator: different from rigid or solid contact environments which are reasonably assumed in early applications in industrial, nuclear and aerospace applications, the slave environment is now more complex and often the objects in contact are much softer in term of stiffness and can not be described by simple elastic model if good teleoperation performance (accurate and transparent) is expected. In this paper, the research of bilateral teleoperation system considering soft environment in recent 20 years has been surveyed for the first time in literature, to the knowledge of the authors. Following the difference in real applications, in this review the definition of soft environment covers linear elastic environment with much lower stiffness than conventional industrial environment and nonlinear complex soft environment with/out time-varying characteristics. Accordingly, the surveyed control strategies and structures in recent literature to improve the stability and accuracy of bilateral teleoperation with soft environment are classified and explained. Finally, the main applications, current challenges and future perspectives of bilateral teleoperation with soft environment are discussed.