Text selection is a common and essential activity during text interaction in all interactive systems. As Augmented Reality (AR) head-mounted displays (HMDs) become more widespread, they will need to provide effective interaction techniques for text selection that ensure users can complete a range of text manipulation tasks (e.g., to highlight, copy, and paste text, send instant messages, and browse the web). As a relatively new platform, text selection in AR is largely unexplored and the suitability of interaction techniques supported by current AR HMDs for text selection tasks is unclear. This research aims to fill this gap and reports on an experiment with 12 participants, which compares the performance and usability (user experience and workload) of four possible techniques (Hand+Pinch, Hand+Dwell, Head+Pinch, and Head+Dwell). Our results suggest that Head+Dwell should be the default selection technique, as it is relatively fast, has the lowest error rate and workload, and has the highest-rated user experience and social acceptance.