Under the background of global climate change, the sustainable development of a society is extraordinarily endangered by the growing frequent and severe flood in China. Flood hazard is a highly uncertain and high-dimensional system. Thus, more difficulties emerge when carrying out flood risk evaluation. In this paper, the Dempster-Shafer (DS) evidence theory was applied to evaluate the flood hazard risk due to its ability in multisource information fusing and information uncertainty processing. The establishment of basic belief assignment (BBA) called evidence model is the nodus and hotspot in DS evidence theory research. In order to solve this problem, BBA was established and combined with comprehensive relative membership degree of index by using variable fuzzy sets theory (VFS). More specifically, the value of BBA was assigned by relatively comprehensive membership degree of VFS theory during the calculation process. BBA of hazard-formative environment, disaster-inducing factors and hazard bearing body were seen as the three pieces of evidence, and these evidences were combined by the classical evidence combination principle. Then, the decisions of flood disaster risk assessment were made by comparing quality function, belief function, relief function and uncertainty of each risk degree. So far, the objective and reasonable evidence model was established in this way while the conflict evidence was greatly reduced. Through evidence combination and the calculation of confidence, likelihood and uncertain interval of different risk degrees based on the classic evidence combination principle, the flood hazard risk was accurately assessed. The DS_VFS model was applied to evaluate the flood hazard risk in Cheng Du where flood disaster has been serious since ancient times. The study area was divided into 12 partitions and each subarea's hazard risk degree was calculated respectively by using DS_VFS model. The results indicated that the risk degree of Cheng Du was high, especially in the northwest, northeast and southwest. According to the spatial and time analysis, the risk growth was mainly due to three factors, increase in population, lagging infrastructure in economic development and continuous deterioration of the natural environment. In order to verify the DS-VFS model, the same historical flood data, altitude and slope data, rainfall data series and socioeconomic data were used in VFS theory, and the result was compared with the flood risk degrees calculated by DS-VFS. The comparison showed that the results of the DS_VFS model were more accurate and accorded better with the actual situation than the VFS results. The multidimensional information of social, economic and natural environment was fused, and uncertain interval of different degrees was calculated. Moreover, the purpose of accurate assessment of flood disaster risk in the study area was achieved by using the DS_VFS model, which was developed from the traditional fuzzy model. However, there were some problems that remain unaddressed in the current study. The conflict evidence was greatly reduced by calculating relative-membership grades, but it was not totally avoided. Besides, wrong decisions can be produced based on the classic evidence combination principle where the conflict evidence was high. In addition, many other calculation methods and decision fusion theories are available to be introduced into the DS_VFS model in the future research, such as support vector machine, absorption method, evidence distance function based on confidence, etc.