The log files generated by digital systems can be used in management information systems as the source of important information on the condition of systems. However, log files are not exhaustively exploited in order to extract information. The classical methods of information extraction such as terminology extraction methods are irrelevant to this context because of the specific characteristics of log files like their heterogeneous structure, the special vocabulary and the fact that they do not respect a natural language grammar. In this paper, we introduce our approach Exterlog to extract the terminology from log files. We detail how it deals with the particularity of such textual data.