The calculation of convective heat transfer coefficients is the key problem of machine tools’thermal analysis. Taking the Y directional screw-nut feed system of a Gantry Machining Center as an example, a method for the calculation of convective heat transfer coefficients was proposed based on a neural network and experimental data. A finite element model of the screw-nut feed system was build as this method’s base, and by using this model’s thermal analytical data, a neural network was build to simulate relationships between convective heat transfer coefficients and the temperature field. Importing measured temperatures into the established neural network, the value of convective heat transfer coefficients was computed. Finally, by comparing results of the above method and a method based on similarity theory with experimental data, the accuracy and validity of this method were proved.