Because of the nonuniformity and instability of atmosphere, it is very difficult to obtain real-timely and precisely a band of infrared atmospheric extinction coefficient with MODTRAN atmospheric transmission software and any other traditional methods. The nonlinear function relationship between the input and output of neural network can be obtained via the relation among the atmospheric optical properties at different wavelengths in the near-infrared band by generalized regression neural network. Then the atmospheric optical properties at other wavelengths can be further derived through real-time inversion of the data at certain wavelength measured by laser radar. The method was proved to be effective in obtainning real-timely the near-infrared atmospheric optical properties of a band in actual combat conditions.