AbstractTests of nonword reading have been instrumental in adjudicating between theories of reading and in assessing individuals’ reading skill in educational and clinical practice. It is generally assumed that the way in which readers pronounce nonwords reflects their long-term knowledge of spelling–sound correspondences that exist in the writing system. The present study found considerable variability in how the same adults read the same 50 nonwords across five sessions. This variability was not all random: Nonwords that consisted of graphemes that had multiple possible pronunciations in English elicited more intraparticipant variation. Furthermore, over time, shifts in participants’ responses occurred such that some pronunciations became used more frequently, while others were pruned. We discuss possible mechanisms by which session-to-session variability arises and implications that our findings have for interpreting snapshot-based studies of nonword reading. We argue that it is essential to understand mechanisms underpinning this session-to-session variability in order to interpret differences across individuals in how they read nonwords aloud on a single occasion.