Aimed at the problem that Fourier transform infrared (FTIR) spectral analysis can’t be used solely to analyze complex gas mxiture, in which there are many gas compositions, some compositions have same molecular group, and the concentrations of some compositions range greatly, in the present paper, a new spectral analysis method is proposed. For this method, the effects of gas temperature and gas pressure on the analysis result are taken into account, and feature variable extraction, recognition and correction of spectral disorder, robust modeling and multi-layer modeling with neural network (NN) are used to build gas mixture analysis models. At the end of this paper, on-line analysis of five alkane gases is taken as example to test the performance of the analysis method. CH4, C2H6, C3H8, iso-C4H10 and n-C4H10 are taken as object gases while iso-C5H12 and n-C5H12 are looked as disturbing gases. The comparison results that analysis result curves of FTIR overlap that of gas chromatograph indicate that FTIR can be solely used in on-line and in-situ analysis of multicomponent gas mixture with same molecular group if the analysis method presented in this paper is used.