Abstract: Conventional gravity field measured on surface reflects the superposition of effects of topographic masses, local gravity anomalies owing shallow geological structures to a regional gravity field caused by deep tectonics. In this paper, we implement different depth layes' regional and residual anomaly separation based on robust 2D polynomial fitting with scalable moving window and weighting function determined by grid distances from surrounding points to central point. The algorithm overcomes the limitations of potential field separation merely in horizontal plane and vertical qualitative research by traditional algorithms. The validity of the algorithm proposed is verified by the data of synthetic models and field measured ground gravity data.