There are many reservoir parameter prediction methods which use seismic attributes. Most of them are statistical methods designed according to a specific parametric model, which means they can be used only in a parametric model. In the actual applications of reservoir parameter prediction using seismic attributes, parametric model may result in great errors if the proper parametric model can not be given on account of the very complicated relationship between reservoir parameters and seismic attributes. This paper has briefly introduced a basic principle and characteristics of the nonparametric regression analysis, and discussed how to analyze and select these seismic attributes which have clear physical meanings, sensitively reflect the lateral variations of reservoir parameters and have good relations with reservoir parameters. The porosity plane distribution has been predicted in Tangjiahe area, the Dagang Oil Field. It has been proved by actual examples that this method will be widely used in the prediction of reservoir parameters and in the characterization of reservoir.