The recovery of information from large databases based on similarity approach supposes a high computational cost when the process is carried out comparing each one of the records with the search pattern. If the database records store some data structure representing the information of the problem domain by means of a graph it is possible to classify these records using a hierarchical model, which considers the structural basic elements of the graphs and diminishes the computational cost of the recovery process considerably. In this paper we propose a classification model based on structural elements (cycles and chains) for large and medium databases.