People’s memory for an event is known to be affected by their verbal descriptions prior to memory assessment. The present experiment investigated whether the computational difficulty of production itself, which is known to affect what people say, can shape descriptions and subsequent event memory. Participants viewed simple scenes and were asked to describe them using either speech or silent gesture, the latter being a much more difficult task. We hypothesized that gesturing participants would over-use action pantomimes, which would yield poorer Inaction (vs. Action) scene memory. Following scene descriptions, participants were given a forced-choice recognition task to discriminate previously presented scenes from foils. Patterns of gesturing showed that gesturers used action pantomimes for Inaction scenes, and they performed reliably worse on Inaction scene memory. Increased production of action pantomimes predicted increased guesses for Action scenes at test, independent of the correct response. Implications for memory and production are discussed.