Atmospheric transmittance is an important parameter in the thermal infrared remote sensing. For the atmospheric transmittance models in the previous paper, taking the transmittance model of the infrared camera(IRS) onboard chinese environment and disaster monitoring satellite(HJ-1B)as an example, the radiance at top atmosphere over waters was simulated, a series of error were brought to the variables in the atmospheric transmittance model, and the temperature from the simulated radiance was retrieved. Given that, the sensitivity of aerosol model, water vapor, visibility and view zenith affecting on retrieved temperature was analyzed. The atmospheric transmittance model was applied to 4 HJ/IRS thermal infrared images and the water surface temperature on April 17, 21, 22 and 25, 2009 was retrieved in lake Taihu, China. The results show that the atmospheric transmittance model within different aerosol models for any channel produce various errors during the temperature retrieving from the remote sensed radiance, that the error is the largest for the advective fog model and is the lowest for the tropospheric model, that the retrieved error of temperature displays the linear trend with the error of variables in the atmospheric transmittance models, the water vapor is the most susceptible for the temperature retrieving, the second for the view zenith and the third for visibility. When the atmospheric transmittance model is applied to HJ/IRS thermal infrared images for retrieving the water surface temperature, the error on April 17 is a little high which the root mean square error(RMSE)and the mean relative error(MRE)are 1.127℃ and 5.75% respectively, while the RMSE on other 3 days is below 1℃ and their MRE is less than 5%. This illustrates that the atmospheric transmittance model have the perfect application accuracy in the thermal infrared remote sensing.