In this paper we test the hypothesis that authorial style depends on genre and find that the stylistic signatures of authors do indeed vary significantly, even when only considering the most frequent function words. We test this hypothesis by comparing the out-of-sample predictions of three different multinomial logistic regression models. We discuss our findings with regard to stylistic variation at the level of words and in the context of existing knowledge about literary genres in the Classical Age. We conclude by advocating adding logistical regression to the stylometric toolbox as a flexible way to investigate how authors' writing varies depending on context.