AbstractWhen observing others' behavior, people routinely infer personality traits from it. Research on the correspondence bias has shown that they draw these inferences even from behavior that is situationally constrained. Moreover, these inferences often happen spontaneously, that is, when people have no intention of forming an impression. The current research investigated whether the same applies for inferences of ideological categories, such as conservative, feminist, or climate change denier. We present a series of five preregistered experiments (N = 1012) employing both direct and indirect measures from the impression formation literature. In Experiments 1 and 2, we found that people deliberately, as well as spontaneously, inferred ideological categories from ideology‐implying behavioral descriptions. In Experiments 3, 4, and 5, we found these inferences to be very robust, in that they persist even when alternative reasons for the ideology‐implying behaviors are presented, suggesting that they are subject to the correspondence bias. We argue that to the extent that people attribute negative valence (affective polarization) or extreme attitudes (perceived issue polarization) to ideological categories, spontaneous inferences of these categories can serve as a precondition for a range of negative interpersonal outcomes related to political polarization.