I will defend a position on levels of selection in the evolution of social behavior that I call “interactionist group selection” and that is somewhere between “broad sense” (quasi)individualist selectionism and trait group based (quasi)multilevel selectionism. I will first make a distinction between a behavioral trait and the mechanism underlying it and argue that they cannot be equated for evolutionary purposes. This matters because the mechanism is what gets selected but the behavior it produces is what it gets selected for, and in the case of social behavior, they can get decoupled in a way that has consequences for the levels of selection. For example, in reciprocal altruism, the behavioral disposition for reciprocity gets selected because it is beneficial for an individual in a particular social context, but that is due to the interaction it creates between individuals participating in the interaction, i.e. the trait group. This does not, however, make the psychological mechanism underlying the reciprocally altruistic behavior evolutionarily altruistic (and this is not an averaging fallacy either). I will argue that the proper interpretation of trait group selection is not that it is a form of group selection in the sense that there are two different levels for fitness beneficiaries but in the sense that there are group traits (the behavioral interactions) that get selected against alternative group traits (e.g. not interacting), and these traits are still dependent on making the individuals participating the interaction fitter than those who are not.