The carbonate fractured-vuggy reservoir is characterized by strong heterogeneity, fast lateral lithology alternation, which brings great difficulties for seismic reservoir prediction. Firstly, this paper used single seismic attribute to select seismic attributes such as time, amplitude, frequency and absorption attenuation which are corresponding with the reservoir well; secondly, carried out correlation analysis and intersection analysis of these attributes, and reserved only one type of the attributes which has high correlation degree to reduce the redundancy; and then carried out cluster analysis and grey correlation analysis between the reservoir parameters and the reserved attribute, and optimized three types of attributes including amplitude alternation, arc length and band width which have high grey correlation degree with carbonate fractured-vuggy reservoir parameters; finally, completed reservoir identification and prediction by 3D cross-plot analysis and neural network analysis.