Water temperature is an essential factor that affects the ecology and environment of the river. Conventional river water temperature (RWT) measuring methods are limited by the geographic location of the hydrological station; thus, RWT data is under a sparse spatial resolution. With the improvement of the spatial resolution of thermal infrared sensors mounted on the satellites within the past twenty years, thermal infrared imagery has gradually been used to retrieve RWT. Currently, most research focuses on using sole sensor and specified processing methods without comparing with other sensors and retrieval approaches. Taking the upper Yangtze River catchment as a case study, this paper evaluates the impact of the following three steps within the entire process on the accuracy of RWT:thermal infrared data (MODIS and Landsat); atmospheric correction methods (the radiation transfer model and JM&S single-channel algorithm); atmospheric correction parameters (Atmcorr and GEOS-5 FP-IT). The results indicated that (ⅰ) MODIS data cannot meet the requirements of the study area due to its low spatial resolution; (ⅱ) As the atmospheric water vapor content is relatively high in the study catchment, the JM&S single-channel algorithm performs poorly on the retrieval; (ⅲ) When there is measured RWT within the study catchment, the proper retrieval process will be using the RTM combined with Atmcorr atmospheric correction parameter and further corrected by using the logistic regression while the final RMSE is about 1.0-1.1℃; (ⅳ) For the basin without measured RWT, the more suitable way is to use the GEOS-5 FP-IT instead and set an atmospheric correction parameter threshold. Its RMSE is approximately 1.2℃. Meanwhile, the results also present that atmospheric transmittance and upward radiation are the key parameters affecting the accuracy of RWT. Therefore, we highly recommended that a similar sensitivity analysis should be applied to other catchments for further studies.