In order to improve the efficiency of GNSS (global navigation satellite system) controlled land leveling system, the feasibility of automatic GNSScontrolled land leveling system was discussed. A path planning algorithm for automatic GNSScontrolled land leveling system was proposed. Firstly the hardware design on early foundation was improved and the tension acquisition module was designed as well as correcting it, which provided data for later path planning. In this method, the global path planning and local realtime path planning were combined to forkliftload or full load the shortest and the longest effective operation of the optimal evaluation, integration and density mean Kmeans clustering farmland grid; then according to the design of low forklift excavation high filling method and local search strategy to give ground path global planning and the actual work was carried out in accordance with local adjustment and planning tension sensor. Farmland automatic navigation ground experiment showed that the maximum elevation difference after the plane leveling went down from 22.8cm to 2.7cm and the height standard deviation of elevation dropped from 12.6cm to 1.5cm; the cumulative percentage, of which the accuracy errors were less than 5cm, was increased from 81% to 97% after the land leveling. This method can achieve better route planning and efficient working of tractors. The full loaded rate and unloaded rate of forklift, the sum of which was about 20%, were less than those of the other methods, increasing the effective working time and providing technical support for automatic GNSScontrolled land leveling system.