A neural network model for estimating the atmospheric transmittance of infrared radiation is established on the basis of the experimental temperature data of a blackbody measured at fixed points. After the temperature of a blackbody is measured at different distances, a BP network is introduced to adaptively learn the underlying regulations of the measured data. The functional relationship between the current measuring distance and the measured temperature is set up. The actual temperature of the target is calculated accurately. The simulation result shows that the built BP network can learn the sample information effectively and the nonlinear atmospheric transmittance model established can solve the problems such as the difficult calculation of atmospheric transmittance of infrared radiation due to the complicated influences.