Abstract: In the middle and late stages of offshore oil field development, prediction and delineation of fluvial sand body boundaries become a key problem to be solved for tapping out remaining oil. Comparing to the large well spacing offshore, fluvial sand bodies are too small to be identified with conventional logging and seismic methods commonly used in the onland areas with dense well networks. Based on the seismic attributes which can reflect the lateral spreading of sand bodies, using the methods of joint bilateral filtering, Otsu adaptive threshold processing and morphological algorithm to improve the Canny operator, a set of adaptive and high-precision prediction methods for fluvial sand body boundaries are proposed in this paper. From the depositional characteristics and evolution pattern of fluvial-facies sand bodies, forward models are established to verify the effectiveness of the method mentioned above. The method applied to the target layer in the Q Oilfield of the Bohai Sea is successful.