This work aims to explore whether and to what extent current machine translation systems are suitable for literary texts. To do so, one stylistic feature of literary texts is examined in particular: free indirect discourse. By comparing French to English machine translation output of free indirect discourse in the novel Thérèse Desqueyroux, by François Mauriac, this three-pronged study provides insight into both the number of errors and the kinds of errors machine translation systems are most apt to make when confronted with certain narrative features. Findings show that of the three systems chosen (Bing, Google, and DeepL), DeepL performs the best. Furthermore, the systems displayed fewer errors in the segments without free indirect discourse. However, a closer look at the translations of free indirect discourse show that the discrepancy in results may not be due necessarily to features of free indirect discourse, and that further research is necessary.