Natural language processing has previously been used with fairly high success to predict a writer’s likelihood of committing suicide, using a wide variety of text types, including suicide notes, micro-blog posts, lyrics and even poems. In this study, we extend work done in previous research to a language that has not been tackled before in this setting, namely Greek. A set of language-dependent (but easily portable across languages) and language-independent linguistic features is proposed to represent the poems of 13 Greek poets of the 20th century. Prediction experiments resulted in an overall classification rate of 84.5% with the C4.5 algorithm, after having tested multiple machine-learning algorithms. These results differ significantly from previous research, as some features investigated did not play as significant a role as was expected. This kind of task presents multiple difficulties, especially for a language where no previous research has been conducted. Therefore, a significant part of the annotation process was performed manually, which likely explains the somewhat higher classification rates compared to previous efforts.