At present, in the seismic reservoir predictions for thin interbeds, attribute analysis and seismic inversion are the main techniques to predict reservoirs by using seismic data. Both of the techniques intrinsically belong to the inverse problem domain and have their own advantages and shortcomings. This paper presents a attribute seismic inversion Based on PNN neural network that can develop the advantages of the techniques mentioned above.Altough PNN neural network that is mathematical interpolation scheme is only use framework of neural network, we can understanding its behavior by mathematical formula to overcome “black box” of BP neural network. The application of the technique obtained a good effect in the sandstone predictions in GTZ Fuyang oil layer, and the identification coincident rate to the sandstone, whose thickness is over 3m, is above 90%.