Response time variability and its changes over time have been interpreted as indicative of levels of knowledge automatization. Predominantly, only declines in variability have been examined over the course of practice and growing second language proficiency. We discuss possible scenarios that may involve increasing, rather than declining variability, such as the establishment of new memory representations. We present repeated-measures lexical decision data on novel words encountered incidentally in a lexical decision task, which show increased variability as familiarity increases. Similar patterns were observed in our reanalyses of data from other novel word learning experiments with task demands different from ours: auditory presentation and a posttest after seven days (Brown & Gaskell, 2014), as well as picture-matching and production tests (Bartolotti & Marian, 2014). Jointly, these results suggest that when no prior representation of a word exists, increased variability may index learning during its initial stages.