Vague quantifiers such as “some” or “several” do not correspond to exact numerical values but are instead understood with respect to their linguistic context and co-present visual information (e.g. Coventry et al., 2010; Moxey & Sanford, 1993). This study investigates the influence of co-speech gestures on the interpretation of “several.” Previous experimental work has shown that people associate larger areas with greater magnitudes (e.g., Henik & Tzelgov 1982, Hurewitz et al., 2006), and the same applies to more expansive gestures (Woodin et al., 2020). In our pre-registered study, participants (N=400) were asked to watch a video of a speaker uttering sentences such as “400 people were at the protest. Several of them got arrested.” The hands moved either outwards or inwards during the underlined phrase (between-subjects). We analysed numerical estimates as a function of gesture with a Bayesian mixed beta-binomial regression model, finding a weak gesture effect: participants assigned higher quantities to a speaker performing an outwards gesture than a speaker performing an inwards gesture (logit coefficient: b=0.07, [-0.09, 0.22]). Confounded factors that became apparent after conducting the experiment led us to pilot a within-subjects follow-up study (30 participants) which additionally contained a no-gesture condition. This design showed a much stronger gesture effect (logit coefficient: b=0.89, [0.56, 1.22]), suggesting that the weak results of Experiment 1 may be because within-subjects control is needed to account for individual differences in how quantifiers are interpreted. We discuss the implications of these findings for future work in numerical cognition and multimodal communication.