Mathematical morphology tools have already been applied to a large range of application domains: from 2d grey-level image processing to colour movies and 3D medical image processing. However, they seem to have been seldom used to process 3D seismic images. The specific layer structure of these data makes them very interesting to study. This paper presents the first results we have obtained by carrying out two kinds of hierarchal segmentation tests of 3D seismic data. First, we have performed a marker based segmentation of a seismic amplitude cube constrained by a picked surface called seismic horizon. The second test has consisted in applying a hierarchical segmentation to the same seismic amplitude cube, but this time with no a priori information about the image structure.