Arable land gradation, a further step of the finished agricultural classification,has a great signifi-cance to the evaluation of arable land quantity and its quality, as well as the protection and utilization of land resources. High resolution QuickBird images are used and a novel strategy different from the traditional gra-dation of farmland is proposed by adopting high-pass filtering fusion method in spatial domain and geometric correction. Accordingly,object interpretation symbols and screen vectorization classification are established to make a real time land utilization map. The effect of large-scale mapping is improved and a map of arable land quality gradation is made at a scale of 1:2000.