Carbonate reservoirs can be extremely heterogeneous with complex rock composition and pore structures, and it is difficult to distinguish the fluid type by conventional reservoir description methods. In this paper, we establish a technical method for the detection of carbonate fractures by the combination of a modified rock physics model and shear wave splitting. Firstly, according to the rock physics characteristics of the Yingshan formation carbonate reservoirs in the S area of the Tarim Basin, we take the various rock matrix components into consideration and modify the Chapman model to establish a multi-scale rock physics model. Secondly, the shear wave seismogram is recorded with different fracture densities and different fluid types using a reflectivity method. Then the shear wave splitting properties are analyzed. We confirm that the amplitude difference between the fast and slow shear wave is quite small when the formation is saturated with brine while there is a large difference between them when it is saturated with gas. Also, we verified that the time delay between fast and slow shear waves increases with the development density of fractures. Thus we can well detect fractured reservoirs using these two attributes of time delay and amplitude difference between the fast and slow shear wave. Finally, this valuable attribute, amplitude difference, as well as the time delay, were both applied to the three-component seismic data of our study area. The prediction results show that fracture sets and gas reservoirs are more developed in the southwest of our study area, which agrees with drilling and trial production data. Hence it proves the reliability of our prediction results and the validity of our prediction method.