ABSTRACT The article attempts to determine whether translation errors (in particular semantic and stylistic ones) in translator-training settings are predominantly the result of translation directionality, i.e. of the fact that the student translators are translating into their L2 and that their language competence in L2 is not as strong as in their L1, or the result of poorly developed thematic and information-mining competences. In order to determine which competence is decisive for the quality of translation output, 112 translations by 14 student translators in the final year of the study at MA level created under experimental conditions were assessed for accuracy (semantic content) and stylistic quality (register and collocations). The students were asked to translate four texts (two on a general, familiar topic and two on a specialized, unfamiliar topic) into their L1 and into L2. The results obtained by using an adapted version of an item-based evaluation method show that factors other than directionality impact translation quality more decisively, among them the level of L2 proficiency and the broadness of the translator’s general knowledge.