Abstract: The spectral signature of a pixel in remotely sensed image in most cases is the result of the reflecting spectral properties of mixed land cover types constituting the area of a pixel.However,despite this phenomenon most remotely sensed image classification algorithms aim at sorting a pixel according to the spectral statistic features of a pixel.Spectral unmixing can not only give the abundance images of surface cover types constituting the area of a pixel,but also get the classification image.In this paper,we process and analyze the TM image of the Yellow River Mouth received on June 25,1999 as the following:(1) Atmospheric calibration of the image data by the internal average relative reflection,(2) Selection of the training pixels of the endmembers,(3) Spectral unmixing of the image data by the logistic model,(4) Getting the abundance image of every endmembers constituting the area of a pixel,and giving the classification image.In the end,the final image resulting from logistic model is compared qualitatively with similar products derived from maximum——likelihood classifier and spectral angle mapping technique.Then the factors effecting the classification product of logistic model are discussed.Moreover,some research aspects for the future are suggested.