How does the visual system recognize human interactions, and what is the nature of these representations? Past work suggests observers can automatically recognize event category (e.g., kicking) and event role information (Agent, Patient) from brief displays, so-called rapid gist extraction of event structure. Questions remain though about how quickly event representations are computed and when they might interface with linguistic/cognitive systems. We explored these issues using linguistically guided visual search. Participants' eye movements were recorded as they heard spoken input (e.g., “The red person is kicking the blue person”) and searched for the matching image. By manipulating visual preview time prior to hearing the critical verb, we can estimate an upper bound of when visually recognized event information is available to ongoing linguistic processes. And by manipulating the posture of the humans in these images, we can help clarify how event representations are recognized and refined over time.