The present paper proposes a spectral discrimination method combining spectral information divergence with spectral gradient angle (SID×tan(SGAπ/2)) which overcomes the shortages of the existing methods which can not take the whole spectral shape and local characteristics into account simultaneously. Using the simulation spectra as input data, according to the interferogram acquirement principle and spectrum recovery algorithm of the temporally and spatially modulated Fourier transform imaging spectrometer (TSMFTIS), we simulated the distortion spectra recovery process of the TMSFTIS in different maximum mix ratio and distinguished the difference between the recovered spectra and the true spectrum by different spectral discrimination methods. The experiment results show that the SID×tan(SGAπ/2) can not only identify the similarity of the whole spectral shapes, but also distinguish local differences of the spectral characteristics. A comparative study was conducted among the different discrimination methods. The results have validated that the SID×tan(SGAπ/2) has a significant improvement in the discriminatory ability.