In a relational database, some tables are used to gather additional information to rows of tables forming the core of the database. This data is stored in tables that we call ``nomenclature tables''. Being able to distinguish them offers many interests in the study, mainte- nance and evolution of databases. We propose properties to define the nature of these tables. Then, an experiment to validate the proposed properties is described and applied on a case study. A classification model for nomenclature tables is built using a datamining algorithm. Its evaluation shows a precision of 88.6% and a recall of 88.7%.