The feature information of ballistic target is deficient in space based infrared system. Spectrum represents the inherent property difference of the object, which can be used as main gist for target recognition. The characteristic spectra of missile plume are taken as an important mean for target recognition. Considering the spectral absorption characteristics and the band extraction rules, we construct a band extraction model using forward and backward interval partial least squares, use to a new spectral adaptive varied weight measure (SAVM) distinguish, and propose a new recognition method based on characteristic plume spectrua, which can be used for the recognition of boost phase ballistic target. The characteristic bands are extracted and SAVM superiority is verified. The method proposed needs less data, provides higher precision and can be a better reference for optimizing infrared detection and recognition system in comparison with the way of using whole spectral range.