The carbonate reservoir is characterized by various types and strong inhomogeneity. Seismic waveform variation characteristic is used to semi-quantitatively identify and predict the distribution of high-quality reservoir. The basic principle, realization process and the key parameters of this technology are analyzed. The seismic response pattern is established in reservoir zone by well-seismic calibration, and the seismic waveform characteristics of different reservoirs are analyzed. Seismic trace waveform of target zone is classified by means of self-organizing neural network. The semi-quantitatively evaluation of carbonate reservoir is finally carried out. According to the prediction, two wells have been drilled, and the drilling result is consistent with the prediction result.