Collective behavior, such as shoaling in teleost fish, is driven by the perceptual recognition of conspecific animals. Because social interactions are mutual, it has been difficult to disentangle the exact sensory cues that trigger affiliation in the first place from those that are emitted by receptive and responsive shoal mates. Here, we overcome this challenge in a virtual reality assay in zebrafish. We discovered that simple visual features of conspecific biological motion provide a potent shoaling cue. Individual juvenile fish shoal for hours with circular black dots projected onto a screen, provided these virtual objects mimic the characteristic kinetics of zebrafish swim bouts. Other naturalistic cues previously implicated in shoaling, such as fish-like shape, pigmentation pattern, or non-visual sensory modalities are not required. During growth, the animals' stimulus preferences shift gradually, matching self-like kinetics, and this tuning exists even in fish raised in isolation. Virtual group interactions and our multi-agent model implementation of this perceptual mechanism demonstrate that kinetic cues can drive assortative shoaling, a phenomenon commonly observed in field studies. Coordinated behavior can emerge from autonomous interactions, such as collective odor avoidance in Drosophila, or from reciprocal interactions, such as the codified turn-taking in wren duet singing. We found that individual zebrafish shoal autonomously without evidence for a reciprocal choreography. Our results reveal individual-level, innate perceptual rules of engagement in mutual affiliation and provide experimental access to the neural mechanisms of social recognition. VIDEO ABSTRACT.