Abstract: The self-organizing neural network technology，combining with seismic attribute，is often used in the automatic identification of seismic facies，but in practical applications，this method has some problems to solve difficultly，such as the classification and recognition of neural network，the selection of seismic attributes and the resolution of the Clustering self-organizing image in an orderly manner and so on.In this paper，some improvement on Kohonen self-organizing networks is made，and the sensitive attribute analysis techniques are used to solve the problem of the choose of seismic attributes，and finally，combining self-organizing network clustering parameters，using of RBF reservoir parameters，the multi-attribute reservoir prediction accuracy is predicted and improved more effectively.

Key words: Seismic phase, self-organizing neural network, seismic attribute, sensitive attribute analysis, reservoir prediction, RBF Network