Influenza exhibits two apparently contradictory features: on the one hand, any given individual can get infected with the disease over and over again, since the virus mutates fast enough to escape acquired immunity; on the other hand, on any given epidemic season, the viral strain is sufficiently well-defined, so that an effective vaccine can be identified. It represents a problem in evolutionary dynamics in which the molecular level and the epidemiological level are tightly bound. Several mechanisms have been proposed to explain this behavior, but a satisfactory answer is still lacking. I shall review the recent debate on the subject and point out the most promising directions of research.