In comparison with instrumental sight reading of musical notation, sight singing is typically characterized by the presence of lyrics. The purpose of this study was to explore how skilled sight singers divide their visual attention between written music and lyrics and how their eye-movement behavior is influenced by musical stimulus complexity. Fourteen competent musicians performed 10 newly composed songs in a restricted temporal condition (60 bpm). Eye movements and vocal performances were recorded and complemented with posttask complexity ratings and interviews. In the interviews, the singers emphasized the priority of focusing on the melody instead of the lyrics. Accordingly, eye-movement analyses indicated not only more total fixation time on music than lyrics but also longer fixation durations, longer durations of visits (i.e., sequences of fixations), and a larger number of fixations per visit on music than on lyrics. The singers also more typically arrived at a bar by glancing first at the music instead of lyrics. Generalized linear mixed-model analyses showed that the number of notes and accidentals in a bar influenced the fixation time and that pupil dilation was increased by a larger number of accidentals. Measurements of eye–voice span, that is, the temporal distance between fixating and singing a note, were best predicted by phrase structure and the note density of previous melodic material. According to the interviews, the best sight singers’ approaches were characterized by a flexibility of moving between different sight-singing strategies. The study offers a comprehensive overview regarding the bottom-up and top-down aspects affecting sight-singing performance.