Abstract: Linear features, such as roads and irrigation channels, make a result of about 12.8% errors in cotton area estimation using Landsat TM images in Northwest China's Xinjing Uygur Autonomous Region. For the operationally monitoring using remote sensing in China Ministry of Agriculture, a sampling method for linear features extraction and measurement to reduce errors was developed by the authors. The theory, approaches and field experiment are introduced in this paper and the results show the method can efficiently improve the results of cotton area estimation using Landsat TM.