The deciding weight theory of Principal Components Analysis is used to compute contribution values of seismic attributes to forecasting parameters, and the attributes sensitivity analysis is solved by getting rid of these attributes with lesser weight coefficient. In this way, the association between reservoir parameters and attributes is established. By means of K-L transform, higher dimensional attributes are mapped to lower dimensional ones while correlation between attributes is eliminated, thus completing the optimization of attributes. In this paper, the target parameters are predicted by BP neural network. The application shows that the dual optimization method combining Principal Components analysis with K-L transform overcomes the limitation of either method and at the same time possesses their respective merits. In short, the method gives a satisfactory solution to the sensitivity analysis, association, and combination optimization of seismic attributes, and eventually improves the precision, speed, and efficiency of reservoir prediction.