This study investigated cognitive aspects of sight translation by analysing the reading behaviour in the process and the output. In our empirical study, two groups of participants—interpreting trainees and untrained bilinguals—carried out three tasks: (a) silent reading, (b) reading aloud, and (c) sight translation. The results show that the two groups were almost identical in the first two tasks, further substantiating the similarity of their language command, but were drastically different in how they tackled sight translation. Interpreting trainees provided much more accurate, fluent, and adequate renditions with much less time and fewer fixations. However, their efficiency at information retrieval was statistically similar to that of the untrained bilinguals. Thus, interpreting trainees were more efficient by being more “economical” during reading, rather than by reading ahead faster, as some would intuitively expect. Chunking skills seem to have also been at play behind their remarkable performance.