In this paper, we present an ongoing project of tagging early French dictionaries, dictionnaire Universel de Basnage de Bauval (1701) in order to build a lexical database which would enable fine-grained queries for linguistic and historical studies. The tagging model is SGML and as a tagset basis we chose to adopt the Text Encoding Initiative Guidelines for print dictionaries. We show that in spite of numerous inconsistencies, computerizing Le dictionnaire Universel is a feasible task and exhibits many regularities in the text which can be partially automated with finite state automata. Applying a systematic grammar on the text renews the metalexicographic studies of early dictionaries.