In order to improve the automation of object-oriented classification，combines the watershed transform and ISODATA clustering to implement a quick and efficient object-oriented classification of remote sensing image. Firstly, uses the improved watershed transform to achieve better segmentation of high resolution imagery; then uses the ISODATA clustering to classify the object segmentation. The experimental results show that this new method obtains better classification effect and fast classification, and greatly enhances the automation of remote image classification.