Abstract: Seismic attributes includes physical attribute and geometry attribute, which quantifies specifically the characters of geometry, kinematics, dynamics or statistics in seismic data. Although geometry attributes can be easily accepted and straightly identified by the sense organ of men, the physical attributes derived from abstract process and mathematics are better than geometry attribute in seismic reservoir prediction. Therefore, seismic attribute is mainly referred to physical attribute calculated by mathematical algorithm. Reservoir prediction by seismic attributes is widely used in geophysics. Since 1980s,pattern recognition technique is paid great attention to, and the reservoir prediction techniques ,such as a fuzzy pattern recognition, statistic pattern recognition, neural network pattern recognition and function approach ,have been successively developed .The predicted objects include hydrocarbon, reservoir thickness, lithology and reservoir porosity. In reservoir prediction, the selection of seismic attribute is accomplished by experience of interpreters, whose effect is subject to better geological conditions, simple predicted objects, and higher S/N in original seismic data. However, under the other conditions, the effect of prediction is worse. In fact, there exist complex relations between predicted objects and their seismic attributes. The seismic attributes sensitive to predicted objects are not totally the same in different areas and reservoirs. They are also somewhat different even for same reservoir and same area. The optimization technique of seismic attributes is an effective means for solving the above questions. The optimum methods of seismic attributes mainly include the dimension-reduced projected profile and cluster analysis etc. Its purpose is to optimize the minimum seismic attributes or seismic attribute esgroup, which are the most sensitive (or most effective, most representatives) to studied problem, in order to increase reservoirs prediction precision and to improve the effect of processing and interpretation related to seismic attributes.