The baseline of an infrared spectrum will drift when a Fourier transform infrared spectrometer is used to detect pollution gases. An infrared spectral baseline correction method based on improved iterative polynomial fitting is proposed in this paper. The method automatically identifies characteristic spectral peaks and subsequently cuts off the characteristic peaks from the spectrum. Then, it uses the bilateral threshold to iteratively fit the infrared spectrum to obtain the best-fitting baseline. The proposed method was verified using the simulated and measured spectra of trichloroethane and ammonia, and the results were compared with those obtained by the three common polynomial fitting methods. The results show that the characteristic peaks of the corrected spectra are more obvious and the baseline is more gradual after baseline correction using proposed method. The simulation and measured data show that our baseline correction method improves the accuracy of the existing polynomial fitting baseline correction method and exhibits good performance.