Abstract In order to understand the development of automatization in second language (L2) acquisition and its role in speaking fluency development, 39 English as foreign language (EFL) learners performed a speaking task (subsequently analyzed for utterance fluency) as well as semantic classification (lexical processing measure) and maze (grammar processing measure) tasks at two time points (Time 1 and Time 2) over six months. The findings indicate that learners significantly speeded up their lexical and grammar processing, as reflected in faster reaction time (RT). However, only lexical processing showed a propensity for more stable processing at Time 1, as indicated by coefficient of variance (CV). Furthermore, multiple regression analyses revealed that more automatized grammar processing (faster RT and smaller CV) at Time 1 significantly predicted a larger reduction in mid-clause pause duration and frequency, respectively. These findings underscore the importance of automatization in grammar processing for developing fluency in an EFL context.