One of the most important skills for a pianist is sight-reading, which is the ability to read an unknown music score and play it. In recent years, research has analysed eye movement during sight-reading. However, the definition of sight-reading has varied. In addition, the participants enlisted as experts in most studies have been music college students. The present study aimed to compare eye movements during sight-reading between experts, teachers at a music college and pianists, and non-experts, music college students studying to become pianists, using an eye tracker. Using easy and difficult music scores for two-handed playing, we investigated whether there were differences in the number of eye fixations, fixation duration, and eye-hand span. The definition of sight-reading in this study is to read a novel music score once without playing the piano, and then to play it while looking at the music score. The results showed that the higher the piano performance rating, the longer the eye-hand span. Areas of interest (AOIs) were defined every two rows, including a treble and bass staff in each music score. We conducted a two-factor repeated measures ANOVA (group × AOI) for each dependent variable to analyse fixation count and fixation duration per eye fixation. There was a significant interaction for the fixation count between groups and AOIs both without and with performance in the difficult score. In experts, the number of eye fixations on the difficult part of difficult score increased compared with other part both without and with performances. By contrast, there was a significant interaction for the duration per eye fixation between groups and AOIs in easy score with performance. The duration per eye fixation in experts was shorter than that in non-experts in the easy score with performance. These results suggest that experts get information through short gaze fixations.