What is the form of natural selection on immune responsiveness? For a population at evolutionary equilibrium, there are two different scenarios. First, it is generally assumed that immune defense has both benefits and costs. If variation in immune responsiveness is due to variation in how individuals trade off these costs and benefits, one would expect immune responsiveness to be subject to stabilizing selection. Second, it is well known that an individual's immune responsiveness is often dependent on its overall condition. If immune responsiveness is condition-dependent, one would expect immune responsiveness to be under positive directional selection. We would therefore expect that the form of natural selection on immune responsiveness depends on the relative magnitude of these two sources of variation: variation in how individuals trade off the costs and benefits of defense, and variation in condition. We measured primary and secondary antibody responsiveness to diphtheria-tetanus vaccine in blue tits during winter and investigated the relationship between responsiveness and survival to the following breeding season. We use responsiveness to these antigens as measures of an individual's ability or propensity to mount an antibody response in case of an infection. Interestingly, different measures of responsiveness were subject to different selective regimes: primary responsiveness to diphtheria was subject to stabilizing selection, whereas secondary responsiveness to tetanus was subject to positive directional selection. In contrast, there was no significant selection on primary responsiveness to tetanus or secondary responsiveness to diphtheria. The finding of stabilizing selection on a measure of responsiveness is evidence that immune defense can incur fitness costs; a central but little-tested assumption of theories of the ecology and evolution of immunological defense. The finding of directional selection on a measure of responsiveness is consistent with the idea that immune responsiveness is condition-dependent, although we cannot rule out the alternative explanation that the population is not at evolutionary equilibrium with respect to this trait.