This invention relates to a method for processing at least two seismic data sets (10, 20), each data set comprising several seismic traces (i, j) grouped by bins (Bi, Bj) and by offset classes (Oi, Oj), the method including the following steps: calculating at least one attribute (a(i, j)) characteristic of a similarity between a first trace (i) of a first data set (10) and a second trace (j) of a second data set (20), selecting or not the first and second traces (i, j) according to a selection criterion applied to the calculated attribute (a(i, j)).