Abstract Rhetorical moves are discoursal units that perform communicative functions in a given genre. They have been manually identified in many previous studies taking a corpus-based approach, and these studies have provided important contributions to discourse structure theories. However, manual analysis has restricted the scale and quantity of texts under investigation and the identification of moves is likely to be influenced by researchers' prior knowledge. In contrast, a corpus-driven approach, by applying automatic computational technology to process a large number of texts, possibly results in a better representation of the moves in a genre and minimises researcher bias in move identification. This study applies one typical corpus-driven approach, a bundle-driven approach, to analyse rhetorical moves of PhD abstracts, a move-intensive genre. The study focused on 5-word sentence initial bundles. Almost all the generated bundles could be identified as move indicators. The majority of the indicated moves aligned with the moves proposed for research article abstracts in previous studies, and one new move, Structure, was identified. This study indicates the potential of a bundle-driven approach in exploring rhetorical moves and in examining the linguistic features of moves by means of lexical bundles. Pedagogical implications for EAP writing are also suggested.