Carbonate reservoir is widely developed in the Tarim Basin of China, where a large number of oil fields are discovered. The reservoir storage spaces in the area are mostly secondary dissolution caves and are characterized as strong heterogeneity. One of the key technical problems is how to accurately calculate the volume of these caves. Influenced by desertification surface and deeply buried reservoir, the seismic data collected from the area has the obvious characters of low signal-to-noise ratio and low resolution. The pre-stack seismic inversion often used in here is AVA inversion using partially stacked data to characterize the carbonate reservoir. Comparing with AVO inversion using single trace, AVA inversion suffers from the loss of lateral resolution when characterizing heterogeneity. Considering the question mentioned above, a physical model is used to analyze the influence of AVO/AVA inversion on heterogeneity. On the one hand, the differences in the lateral resolution of AVO/AVA inversion are quantitatively studied. On the other hand, the results of seismic inversion are compared with the actual size of caves in the physical model. The results demonstrate that the signal-to-noise ratio is able to be improved by AVA inversion while the lateral resolution is subsequently reduced. For the AVA inversion, seismic incident angle gather is divided into three parts, using for pre-stack inversion, the reservoir heterogeneity resolution is able to be reduced up to 33%.