Sight translation has been widely used in aptitude testing to screen prospective trainee interpreters at leading interpreter training schools, including ESIT, ISIT, and EMCI. However, it has also been criticised for its lack of validity and reliability. No empirical study has thus far been conducted to explore its power to predict interpreting performance. To fill this gap, the present study adopts a data-driven approach to explore the predictive validity of sight translation for interpreting performance in both consecutive and simultaneous modes. A total of 33 graduate trainee interpreters (A language: Chinese; B language: English) enrolled in an interpreting programme of a leading Chinese university were tracked over the course of one and a half academic years. Two raters (A language: Chinese; B language: English) were asked to assess their performance based on the criteria of fidelity, language, and delivery. Statistical analyses suggest that, irrespective of language direction, the participants' performance in sight translation was not predictive of their performance in either consecutive or simultaneous interpreting. This study provides much-needed empirical evidence for the development, administration, and assessment of aptitude testing for interpreting and interpreter training at large.