Crowd simulation addresses algorithmic approaches to steering, navigation, perception, and behavioral models. Significant progress has been achieved in modeling interactions between agents and the environment to avoid collisions, exploit empirical local decision data, and plan efficient paths to goals. We address a relatively unexplored dimension of virtual human behavior: thermal perception, comfort, and appropriate behavioral responses. Thermal comfort is associated with the ambient environment, agent density factors, and interpersonal thermal feedback. A key feature of our approach is the temporal integration of both thermal exposure and occupant density to directly influence agent movements and behaviors (e.g., clothing changes) to increase thermal comfort. Empirical thermal comfort models are incorporated as a validation basis. Simple heat transfer models are used to model environment, agent, and interpersonal heat exchange. Our model’s generality makes it applicable to any existing crowd steering algorithm as it adds additional integrative terms to any cost function. Examples illustrate distinctive emergent behaviors such as balancing agent density with thermal comfort, hysteresis in responding to localized or brief thermal events, and discomfort and likely injury produced by extreme packing densities.