This is a corpus-based study that investigates instances in which court interpreters in Hong Kong deviate from using direct speech and the first person, notwithstanding the requirement to use both of these when rendering statements made by witnesses or defendants. Quantitative data indicate that court interpreters do adhere to this requirement when interpreting Cantonese into English, but deviate from it when interpreting English into Cantonese. These data suggest that the use of reported speech and/or of the third person has identification functions that help Cantonese-speaking witnesses and defendants follow court proceedings and serve the pragmatic function of adding illocutionary force to interpreted utterances. Data from interviews with interpreters and legal professionals suggest that some latitude is exercised and tolerated when interpreters deviate from using direct speech and/or the first person when the target language is Cantonese. The findings indicate that court interpreters in the corpus observe strict professional guidelines by using direct speech most of the time, but occasional deviation from the direct approach suggests that court interpreters are able to make discretionary decisions to facilitate communication.