How does characterizing a group as hostile and dangerous shape behavior? We present two high-powered experimental studies, a close and a conceptual replication of the ‘Police Officer's Dilemma’ (Correll et al., 2002). Experiment 1 (N = 164)—a close replication—uses the original shooter task with Arab-Muslim targets. Participants showed a so-called shooter bias: A significant interaction in reaction times with faster ‘shoot’ responses for armed Arab-Muslim targets compared to armed White targets (ηp2 = .11, 90% CI [.04; .18]). This provides evidence that the shooter bias is robust against context variations. Experiment 2 (N = 165)—a conceptual replication and extension—investigates whether this effect generalizes to other threat-related behavior. In a novel ‘avoidance task’ with Turkish and White German targets, participants ‘avoid’ armed targets carrying knives and ‘approach’ unarmed targets carrying innocuous objects. Again, we observed a significant interaction effect: Reaction times were faster for armed Turkish targets, but slower for unarmed Turkish targets as compared to White German targets (ηp2 =.19, 90% CI [.11; .27]). Results are interpreted as an avoidance bias—an effect almost twice as large as in the original shooter task. We discuss that the avoidance task may be cognitively more demanding than the shooter task and that the avoidance task may provide a more subtle measure of bias in threat detection. This may lead participants to exert less behavioral control. Taken together, this research highlights that threat stereotypes have powerful influences on judgment and behavior, with the potential to disrupt intergroup interactions.