Mammalian predator-prey systems are behaviorally sophisticated games of stealth and fear. But, traditional mass-action models of predator prey dynamics treat individuals as behaviorally unresponsive “molecules” in Brownian motion. Foraging theory should provide the conceptual framework to envision the interaction. But, current models of predator feeding behavior generally envision a clever predator consuming large numbers of sessile and behaviorally inert prey (e.g., kangaroo rats, Dipodomys, collecting seeds from food patches). Here, we extend foraging theory to consider a predator-prey game of stealth and fear and then embed this game into the modeling of predator-prey population dynamics. The melding of the prey and predator's optimal behaviors with their population and community-level consequences constitutes the ecology of fear. The ecology of fear identifies the endpoints of a continuum of N-driven (population size) versus μ-driven (fear) systems. In N-driven systems, the major direct dynamical feedback involves predators killing prey, whereas μ-driven systems involve the indirect effects from changes in fear levels and prey catchability. In μ-driven systems, prey respond to predators by becoming more vigilant or by moving away from suspected predators. In this way, a predator (e.g., mountain lion, Puma concolor) depletes a food patch (e.g., local herd of mule deer, Odocoileus hemionus) by frightening prey rather than by actually killing prey. Behavior buffers the system: a reduction in predator numbers should rapidly engender less vigilant and more catchable prey. The ecology of fear explains why big fierce carnivores should be and can be rare. In carnivore systems, ignore the behavioral game at one's peril.