The existence of mixed pixels is the main factor influencing the classification accuracy of remotely sensed image. Fuzzy classification is an important method of unmixing the mixed pixels. Its results depend on how accurate the membership value to various types of each pixel after classification corresponds to its actual component. If the clustering number is not equal to the actual type number in the unsupervised classification

or there are some types untrained in the supervised classification

the accuracy of the popular algorithm

namely Fuzzy c-means （FCM） will be degraded. Fortunately

Possibilistic c-means （PCM） is insensitive to it and can work well. This paper proposes the pixel unmixing method of remotely sensed image based on PCM algorithm. The priority of the PCM is illustrated by an actual example in the supervised classification in this paper.