We report on an experiment to investigate the top-down effect of exogenous social identity cues on a multiple-identity tracking task, a paradigm well suited to investigate the processes of binding identity to spatial locations. Here we simulated an eyewitness event in which dynamic targets, all to be tracked with equal effort, were identified from among a "crowd" of 8 faces, as an assailant, bystander, policeman, and victim. Even in such a simplistic paradigm, where no actual assault was witnessed and no consequences were associated with the task, results demonstrated a significant attentional bias, namely that participants were significantly better at tracking the assailant, bystander, and policeman than they were the victim. We found no difference in accurate recall based on the use of text or face cues and no systematic pattern of response errors.