Abstract:T-X and F-X prediction filtering methods are common noise attenuation techniques. These methods are effective to noise attenuation,but they bring some false seismic events and attenuate signal energy at the place of high amplitude noises.The negative effect results from the fact that there is noise in input data. Then inversive prediction filtering is taken to remove noises. In other words,it is used to achieve signal estimation,from which we derive a new filtering factor so as to make a new prediction filtering under less noise influence.There often exist coherent linear noises(such as surface wave,refraction wave etc.)in seismic data which have undergone prestack noise elimination. On the contrary,these noises can be enhanced in prediction noise elimination. The joint use of inversive prediction filtering and operator extrapolation will bring quite desirable result. The joint use of inversive prediction filtering and operator extrapolation preserves useful reflection amplitudes but removes strong coherent noise and the false seismic events which result from prediction filtering. Such effect has been seen in theoretical and real data processing results.