【Objective】To design a GPS-based paddy field leveller and to improve the paddy field leveling and accuracy. 【Method】 GPS receiving antennae were fixed on the top of leveling scraper. The GPS data of dynamic height positioning was used as the height of the leveling scraper. The amplitude limit and recurrence average filtering algorithm and PD control algorithm were applied to control the accuracy and stability of GPS dynamic height positioning. 【Result and conclusion】 The relative standard deviation reduced from 15.8 cm to 4.7 cm. The sampling point cumulative percentage of the absolute difference value no more than 5 cm was over 85.4%. The amplitude limit and recurrence average filter algorithm reduced the error of GPS dynamic height positioning. The accuracy of GPS dynamic height positioning can meet the precision requirement for paddy field leveling. The surface smoothness of paddy field can be obviously improved.