To analyze and extract feature parameters of a temperature field, a sup port vector regression based numerical analysis method is proposed. First, the temperature field of a studied object is modeled by the finite element analysis ( FEA). Furthermore, position and temperature of every node are extracted. Then, the studied object is divided into several feature areas, by which a sensitive di rection to be analyzed is determined. Finally, the support vector regression is repeatedly implemented on every feature area, and its feature coefficients are extracted. Consequently, a allround quantitative evaluation on the heat transfer performance of the studied object is presented. In order to verify the method, a simulated analysis is made on the cooling grid of a heat exchanger. The result indicates that the proposed method is effective in numerical representing a com plex temperature field, which implies its great potential in many fields such as industry process automatic control and mechanical structure optimal design, etc .