Abstract: In this article, a way of dividing whole remote sensing image into subimage has leen developed to classify the type of vegetation. There are different types of vegetation in different geographic unit, so according to geographical region feature in a remote sensing image, the whole image is divided to severl subimages, the types of vegetation in a subimage is less than in a whole image. Types of vegetation in a region graphic is easy to be showed and the threshold of classifying has a wide scape. With the assistance of Digital Elevation model (DEM) and slope maps, classification accurateness was improved. This way was applied in north Xinjiang, the paper hold NDVI (Normalized Difference Vegetation Index) image as data resource, divided whole research area into four regions：Altai grassland region, West Jungger region, Zhaosu region and East Tianshan region. Different classifying standard was used in different region, and then using GIS software together four regions into a whole classified image. Because this method was grounded on knowledge, the classification result of vegetation in Xinjiang north was improved to 89.33%.

Key words: vegetation classification, remote sensing, divided area, North Xinjiang