We present clustering-based methodologies used to process 3D seismic data. It firstly replaces the volume voxels by corresponding feature samples representing the local behavior in the seismic trace. After this step samples are used as entries to clustering procedures, and the resulting cluster maps are used to create a new representation of the original volume data. This strategy finds the global structure of the seismic signal. It strongly reduces the impact of noise and small disagreements found in the voxels of the entry volume. These clustered versions of the input seismic data can then be used in two different applications: to map 3D horizons automatically and to produce visual attribute volumes where seismic faults and any discontinuities present in the data are highlighted. Concerning the horizon mapping, as the method does not use any lateral similarity measure to organize horizon voxels into clusters, the methodology is very robust when mapping difficult cases. It is capable of mapping a great portion of the seismic interfaces present in the data. In the case of the visualization attribute, it is constructed by applying an auto-adaptable function that uses the voxel neighboring information through a specific measurement that globally highlights the fault regions and other discontinuities present in the original volume. We apply the methodologies to real seismic data, mapping even seismic horizons severely interrupted by various discontinuities and presenting visualization attributes where discontinuities are adequately highlighted.