Conventional seismic exploration method based on post-stack data usually fails to identify the distribution of fractured and caved carbonate reservoirs in the Tarim Basin, so the rich pre-stack information should be applied to the prediction of carbonate reservoirs. Amplitude-preserved seismic data processing is the foundation. In this paper, according to the feature of desert seismic data (including weak refl ection, fast attenuation of high frequency components, strong coherent noises, low S/N and resolution), a set of amplitude-preserved processing techniques is applied and a reasonable processing flow is formed to obtain the high quality data. After implementing a set of pre-stack amplitude-preserved processing, we test and define the kernel parameters of amplitude-preserved Kirchhoff PSTM (pre-stack time migration) and subsequent gathers processing, in order to obtain the amplitude-preserved gathers used to the isotropic pre-stack inversion for the identifi cation of caved reservoirs. The AVO characteristics of obtained gathers fit well with the synthetic gathers from logging data, and it proves that the processing above is amplitudepreserved. The azimuthal processing techniques, including azimuth division and binning enlargement, are implemented for amplitude-preserved azimuthal gathers with the uniform fold. They can be used in the anisotropic inversion to detect effective fractures. The processing techniques and flows are applied to the field seismic data, and are proved available for providing the amplitude-preserved gathers for carbonate reservoir prediction in the Tarim Basin.