In this work we present a clusterization-based method to map seismic horizons and faults from 3D seismic data. We describe a method used to quantize an initial seismic volume using a trained instance of the Growing Neural Gas (GNG) algorithm. To accomplish this task we create a training set where each sample corresponds to an entry volume voxel, retaining its vertical neighboring information. After the training procedure, the resulting graph is used to create a quantized version of the original volume. In this quantized volume both horizons and faults are more evidenced in the data, and we present a method that uses the created volume to map seismic horizons, even when they are completely disconnected by seismic faults. We also present another method that uses the quantized version of the volume to map the seismic faults. The horizon mapping procedure, tested in different volume date, yields good results. The preliminary results presented for the fault mapping procedure also yield good results, but needs further testing.