The classification accuracy is unsatisfactory in the complicated terrain area of the Loess Plateau when the single supervised classification is used in the remote sensing investigation of slope field. Taking the loess hill and gully area of northern Shaanxi Province as a test area, the slope field and other categories are extracted by the improved classification which is the integration of supervised classification and unsupervised classification. The result shows that this method improves classification accuracy greatly, and is suitable for the slope field investigation in the loess hill and gully area. The research is of critical significance in obtaining the up-to-date information of the slope field and converting farmland into forest or grass land in slope land reasonably.