This paper proposes a new method of PM2.5 concentration inversion based on difference index through analyzing the spectrum characterization which is sensitive to the particular PM2.5. In addition, the spectral curves of typical land culture vegetation and soil are measured using Avafield-1 spectrometer (range of measurement 300～1 100 nm) under different PM2.5 concentration. The result showsthat the PM2.5 makes the reflectance of vegetation and soilincrease in red band and decrease in near-infrared. Therefore, the difference index (difference, index, DI) of sensitive red and near-infrared is used to characterize the particlesconcentration. This paper uses TM image to obtain the difference index and the PM2.5 concentration inversion of Beijing with the measured dataprovided by the ground air quality monitoring station in Beijing and surrounding areas. The result of the fitting analysis shows that the accuracy (r=0.796) of DI model onMar 1st(average PM2.5=105.8) is higher than that (r=0.628) on Dec14st (average PM2.5=15.8), namelythe accuracy of the DI models is lower when haze pollution degree is weak because the spectral characteristics change caused by particles are not obvious with low particulate content. On the other hand, the quality of remote sensing image is poor when the haze pollution degree is too serious, so this method is suitable for particle concentration inversion under mild and moderate haze pollution. The new method can obtain high spatial resolution (30 m) result with very simple inversion process. In addition, this method can obtain the PM2.5 concentration distribution of different temporal and spatial resolution just by selecting different remote sensing imagescontaining sensitive band data. The new method has wide application prospect.