Collaboration enhances conceptual learning with multiple representations. However, learning with multiple representations also involves perceptual learning processes. These often-overlooked learning processes are the target of perceptual trainings, which expose students to short nonverbal tasks so that students can induce visual patterns across representations. Given the focus of perceptual trainings on nonverbal learning, we investigate the impact of collaboration via gestures without allowing students to talk. On the one hand, gesture-based collaboration may be effective because a partner's gestures may direct students to meaningful visual features. On the other hand, gesture-based collaboration might be ineffective because gesturing may trigger verbal thought, which has been shown to detract from perceptual processing in prior research on the verbal overshadowing effect. We investigated this question in a quasi-experiment with N = 438 chemistry undergraduate students. Students either worked on a perceptual training individually or collaborated using only gestures. Posttest data show an advantage of students working individually. Mediation analysis based on log data revealed a positive mechanism of collaboration enhancing learning gains by reducing students' errors during the training. Gesture analysis showed that students used gestures to nonverbally explain their thinking and that representational gestures reduced error rates whereas other types of gestures did not. This might have detracted students from perceptual processing of the stimuli, creating a "nonverbal overshadowing" effect analogous to the verbal overshadowing effect. Altogether, our findings identify boundary conditions of the benefits of collaboration while also revealing possible pathways for future research to explore perceptual learning in social situations.