Abstract: Based on remote sensing data of land use in 1986, 1995 and 2000 years on Huaxi district at Guiyang city in Guizhou province, spatial autocorrelation of land use changes at scales of 0.2 km×0.2 km, 0.4 km×0.4 km and 0.6 km×0.6 km were studied using GIS technology. Then, space lag model and classical linear regression model at three scales were established. The results showed that cultivate land, forest land, grassland, construction land and water area in Huaxi distrct at three scales in 1986, 1995, and 2000 years had strong positive spatial autocorrelation, the degree of spatial correlation had some connection with the scale, and spatial autocorrelation generally decreased when the study scale became larger. Compared with the classical linear regression model at three scales, the residual of space lag model at three scales was smaller , and better in explain ability and forecast accuracy. The study provides a foundation for land use management for Karst region.