In this paper, we study how existing natural language processing tools for Italian perform on ancient texts. The first goal is to understand to what extent such tools can be used “as they are” for the automatic analysis of old literary works. Indeed, while NLP tools for Italian achieve today good performance, it is not clear if they could be successfully used for the humanities, to support the critical study of historical works. Our analysis will show how tools’ performance systematically vary across different time periods, and within literary movements. As a second goal, we want to verify whether or not simple customization methods can improve the tools performance over the old works.