The thermal error of a machine tool can be regarded as an important index of machining accuracy. Such an error is reproducible and stable, meaning that machine tools can be easily maintained by thermal compensation, ensuring a long-term good machining quality. In this study, reliability engineering was applied to increase the mean time between failures (MTBF) of a cooling system to maintain the structure temperature and accuracy, and the MTBF was increased from 3787 h to more than 23041 h. In addition, the Taguchi method was used with the finite element method and multivariate regression analysis to obtain the optimal cooling conditions to reduce thermal deformation and stabilize the structure temperature. The variation of the temperature of the machine tool structure was improved from ±0.517 to ±0.367 ℃, a reduction of 29%.