To improve the compensation accuracy and robustness of a thermal error compensation model for Computer Numerical Control(CNC) machine center in practical engineering applications, a selection method for the optimal speed of CNC machine center was presented. For the Z-axis of Leaderway V-450 CNC machining center, several batches of experiments were carried out at a variable speed spectrum and a constant speed by controlling the spindle in an idling condition. Then, fuzzy clustering method combined with the gray correlation was used to select the temperature sensitive point and establish multiple linear regression models. Finally, by analysis of the model predictive effect in different type speeds and comparison of the model predictive effect in the same speed type, a selection method for the optimum spindle speed of CNC center in modeling of thermal error compensation was given. The research results show that thermal error compensation models established by different spindle speed regimes based on the international standard have great differences in the thermal error predictive effect of machine tools. It suggests that the compensation model based on optimum spindle speeds selected by practical engineering application will provide a good prediction effect.