Thermal distortion is a critical issue in machine tools, especially in high speed machining. This is the reason why recent machine tools have thermal compensation function. To compensate thermal distortion, it is necessary to make a model that has some relationship between temperature and deformation. Various experimental methods ye widely been used in thermal test: constant spindle speed, unit step speed increase, random spindle speed, etc. This paper focuses on which type of spindle operation condition is good for thermal experiment. Also, experimental data is modeled using multiple linear regression models and compared each other to select a method. Consequently, it turned out at e condition of 75％ constant of maximum spindle speed is good enough to generate temperature and distortion data.