During the seismic interpretation process, geoscientists rely on their experience and visual analysis to assess the similarity between seismic sections. However, evaluating all of the seismic sections in a 3D survey can be a time-consuming task. When interpreters are working on a data set, a common procedure is to divide the cube in increasingly finer grids until they are satisfied with the result of the interpretation. We have developed a method based on graph theory and image texture in which we represent a seismic data set as a complete weighted undirected graph-which we call a seismic graph. The vertices of this graph represent the seismic sections, and the weight of the edges represents the distance between the texture feature vectors of the vertices they connect, allowing for a powerful yet concise representation of potentially large data sets. We have investigated the potential of graph analysis to build an adaptive grid that is more likely to capture the underlying structures present in a survey, providing a tool for a faster and more precise interpretation. The main idea is that such a grid would be finer in regions with more geologic variations and coarser otherwise. To demonstrate the capabilities of our technique, we apply it on a public data set called Netherlands F3. Using our method, we suggest which seismic sections-key sections-should be considered in the interpretation process. The results of our experiments indicate that our methodology has great potential to aid the seismic interpretation process.