Temperature fluctuations induced by incomplete mixing of hot and cold fluids at T-junction used in various industrial plants may cause thermal fatigue failure. The conventional method of evaluating thermal fatigue produces a greatly varying margin. CFD/FEA coupling analysis is expected as a useful tool for evaluating thermal fatigue more accurately. The present authors have established high-accuracy LES prediction methods of fluid temperature fluctuations at T-junction. This study aims to investigate the effect of fluid-structure thermal coupling methods on CFD prediction accuracy of structure temperature fluctuations (STF). For fluid-structure thermal coupling, the wall functions based method and thermal conduction based method were investigated for the coarse mesh and fine mesh, respectively. The LES results show that the fluctuation range and dominant peak frequency of power spectral density (PSD) of fluid temperature predicted by both methods agree with the experimental results well. On the other hand, the fluctuation range and dominant peak frequency of PSD of structure temperature predicted by the thermal conduction based method are close to the experimental results. However, the range of STF predicted by the wall functions based method is just about 55% of that predicted by the thermal conduction based method and the experimental results, although the dominant peak frequency of its PSD is predicted well. Therefore, the thermal conduction based method for a fine mesh is capable of predicting thermal fatigue loadings at T-junction with high accuracy, but the wall functions based method for a coarse mesh greatly under-predicts the thermal fatigue loadings.