The aim of this paper is to use topic modeling method to detect semantic differences between canonical and non-canonical literary works in longer Slovenian prose published between years 1836 and 1918. The analysis is carried out on the Corpus of Longer Slovenian Narrative Prose (KDSP 1.0) by using the Gensim Python library. The results have shown that only three groups of topics (bourgeoisie, rural and uninterpretable) represent more than 75% of topics in the canonical works, while topic distribution in non-canon is less homogenous and shows a lesser preference for the above-mentioned topics. A closer look at the topic distribution within individual texts has shown that canonical and non-canonical works on average contain a similar number of topics, however, above 90% of non-canonical works include one dominant topic that scores above 50%, while 60% of canonical works contain one dominant topic. Moreover, the results have shown that all works by Ivan Cankar, the most canonical Slovenian prose writer, which are included in the KDSP corpus, contain a topic that is specific to Cankar’s texts and that his works are distinctively mono-topical compared to other canonical authors.