Abstract:3D random noise attenuation (3D RNA),a new 3D noise removal,is put forward here.Its presumptive prerequisite is that partial reflection event is in plane so that the signals at the same frequency are predictable in all directions in(f-x,y)domain.As a result,noise removal can be achieved by determining a rectangular predictive filtering operator with the use of the least-square principle of multitrace complex number.Theoretical and real processing data show that this 3D method is much better than usual predictive noise attenuation(2D RNA)in f-x domain.This method is very effective on noise attenuation because it makes full use of 3D information and shows cbmplete geological structure.Such noise removal brings the seismic data which exhibit clear fault points and spontaneous wave forms,without coherent earthworm phenomenon.