The advantages,disadvantages and applicability of different spectral selection models are analyzed. Based on this,a band selecting method based on wavelength growing is proposed. Band selection is divided into two parts:center wavelength and band width. Firstly,the optimal center wavelength is determined based on the visual contrast before target background enter the pupil. And then the appropriate bandwidth is selected to achieve the optimum synthesis signal to noise ratio. In specific implementation,the optimal detection waveband is obtained by iterative growth of center wavelength,and an improved method is proposed for the cases where the general region growth may fall into local optimal solution. By repeating growth on multiple seed wavelengths,the maximum continuous union of local solution is chosen as global waveband. The method is verified in typical missile early warning scene. Influences of target radiance,clutter,detector noise and spatial resolution on spectral selection are analyzed,and the optimum bands for complex backgrounds were obtained at 2.608~2.915 μm and 4.193~4.464 μm. The improved method in this paper has good convergence and high precision.