Data-driven learning (DDL) is particularly well suited to translator education, providing an adequate pedagogic approach for two related yet distinct learning objectives: the development of translation competence and the learning of foreign languages. Interest in data-driven learning in the translation classroom started in the 1990s, as an obvious development of corpus-based translation studies and corpus-assisted language learning. Two main types of translation-driven corpora have been used so far in this setting: bilingual comparable corpora (sets of nontranslated comparable texts in two languages) and parallel corpora (sets of original texts and their translations, aligned at sentence level). While the former raise challenges at the level of analysis, due to the difficulties involved in establishing cross-linguistic equivalences, the latter are easier to consult, but the evidence they offer should be carefully evaluated, as it may be influenced by the translation process. Attempts at assessing the effectiveness of DDL in translator education are still limited. Evidence obtained by comparing translations carried out with and without corpora remains inconclusive, even though a positive learning effect seems to exist, particularly for translation into the foreign language. The perceptions of students are also mixed, due to the cognitive burden of corpus analysis and its time-consuming nature. Yet a pedagogy that favours the acquisition of research skills and flexibility, like DDL, would seem to be particularly well suited at a time when the translation profession is faced with important societal and technological challenges. To address them, educators should favour a change of attitude in students, so that they embrace complexity.