In current remote sensing image segmentation threshold determination method ignores the foreground and background of inner link, lead to segmentation and contour fuzzy, causing poor performance as a whole.For this, put forward a new kind of remote sensing image segmentation based on K - means algorithm, and a method for determining the threshold value of by K - means algorithm for remote sensing image segmentation, the between-cluster variance method are used to get the initial threshold segmentation of remote sensing image based on the threshold will remote sensing image was divided into two categories, the average of two types, as K - means clustering algorithm of two initial clustering center, through the K - means clustering method of successive iteration, constantly updated clustering center, until get the clustering criterion function, thus the optimal threshold segmentation of remote sensing images.The experimental results show that the proposed method is adopted to define in the remote sensing image segmentation threshold, not only high efficiency, and overall superior performance;Will get the threshold value of application in remote sensing image segmentation, can separate the target and background is effective, and the target after separation section outline clearer.