Abstract: To detect and identify pollutant gases in the distance speedily and accurately with IR remote sensing spectrometer,it is necessary to remove high-frequency noise and low-frequency baseline,of which the purpose is to extract feature information. For the deficiencies of the existing methods, EMD was proposed to preprocess IR remote sensing spectrum. EMD is a preprocessing algorithm which works self-adaptively and without parameters. After High-frequency noise and low-frequency baseline was removed, the global assessment factor RMS1, the partial assessment factor RMS2, and the comprehensive assessment factor RMS* reached 0.141, 0.182 and 0.026 respectively. The perfomace of EMD is better than wavelet decomposition method obviously.The result shows that EMD is convenient and reliable for denoising and baseline correction of IR remote sensing spectrum.