There is widespread agreement that social and individual learning are adaptations to varying environments. However, existing theory assumes that organisms cannot detect changes in the environment and instead adapt to averages. This paper develops the first analytical model that allows for the simultaneous coevolution of socially learned traditions, reliance on social learning, and signal detection for environmental change. There are numerous conditions under which detection can be stable once common but cannot invade the population when rare. When signal detection is maintained by selection, it always leads to pure separating equilibria at which organisms always learn individually when they believe the environment has recently changed and otherwise always learn socially. Detection can increase mean fitness at equilibrium, but it may also reduce it.