Web site clustering consists in finding meaningful groups of related web sites. How related is some web site to another is a question that depends on how we represent web sites. Traditionally, vectors and graphs have been two important structures to represent individuals in a population. Both representations can play an important role in the web area if hyperstructure is considered. By analyzing the way web sites are linked, we can build vectors or graphs to understand how a web site collection is partitioned. In this paper, we analyze these two models and four associated algorithms: k-means and self-organizing maps (SOM) with vectors, simulated annealing and genetic algorithms with graphs. For testing these ideas we clustered some web sites in the Central American web. We compare the results for clustering this web site collection using both models and show what kind of clusters each one produces.