To explore the law of land use change and driving mechanism in loess hilly region, based on Landsat TM remote sensing images in 1986 and 2001, distributions of land use patterns were analyzed, and distribution of land use spatial patterns in 2016 had been forecast by using the CA-Markov model. The results showed that land use structure was mainly dominated by slope land before 1986, which occupied 59.67% of the total area, secondly was terrace, grassland and forest. The forest land and the terrace had increased obviously from 1986 to 2001, while slope land reduced evidently, and other land uses had no changes. With regard to single land use type dynamic, the land use type that increased markedly in area had high dynamic degree simultaneously, therefore, the dynamic degree of terrace was the biggest, which achieved to 15.09, the second was forest land. By using CA-Markov model, the predictive results indicated that the land use changes during 2001–2016 period were similar to that during 1986–2001, that is, in general slope farmland would continue to reduce in some degree, while forest land and terrace kept increasing, but the slope farmland reduced slowly compared to that from 1986 to 2001. This is mainly because other slope farmland are restricted by a series factors such as site conditions and the special policies, with regard to other land uses, the mutual conversions present a stable dynamic change for their relatively small area. Land use and land cover pattern can be adjusted according to the simulation results, which may serve as a scientific basis for land planning and management.