This study investigated whether the threat-superiority effect in a visual search task was based on the processing of shapes (lower level), or the processing of semantic characteristics (higher level). Participants were asked to determine whether all pictures in a visual search display were members of the same category (all distractors), or whether there was a discrepant member (the target). The threat-superiority effect was determined by a shorter reaction time for detecting threatening, than non-threatening targets. If the threat-superiority effect was based on semantic processing, it was expected that the effect would be observed even when a target and distractors shared a basic shape, such as a knife among combs, or a comb among knives. Experiment 1 showed a threat “shape”-superiority effect, but not a threat “semantic”-superiority effect. In Experiment 2, when the threatening target was a gun (Japanese participants have no direct experiences with guns, but know that guns are threatening), the threat-superiority effect was not found. These findings suggest that the threat-superiority effect is not based on semantic processing, but on processing of shape characteristics.