In order to reduce thermal errors and improve machine accuracy, a new kind of thermal error modeling method, which takes temperature and machining parameters as input parameters, was established on the basis of time-series regression analysis and numerical computation. The model consists of a thermal error model, a thermal properties model of hydrodynamic bearing and a thermal transfer model of the spindle. Firstly, the correlation model between the temperature data serial of thermal points and thermal error data serial was established on the basis of time series correlation analysis. Then, the number of thermal points was reduced by calculating the grey correlation. In the meantime, the thermal properties model was established based on time-series regression analysis and numerical computation method. A predicting model of thermal error was established for a large CNC gantry grinder. It is indicated by experiment that the thermal error model has good recognition function for thermal error.