Theoretical studies suggest that the abrupt and substantial changes in the productivity of some fisheries species may be explained by predation-driven alternate stable states in their population levels. With this hypothesis, an increase in fishing or a natural perturbation can drive a population from an upper to a lower stable-equilibrium population level. After fishing is reduced or the perturbation ended, this low population level can persist due to the regulatory effect of the predator. Although established in theoretical studies, there is limited empirical support for predation-driven alternate stable states in exploited marine fish populations. We present evidence that egg predation by haddock (Melanogrammus aeglefinus) can cause alternate stable population levels in Georges Bank Atlantic herring (Clupea harengus). Egg predation by haddock explains a substantial decoupling of herring spawning stock biomass (an index of egg production) from observed larval herring abundance (an index of egg hatching). Estimated egg survival rates ranged from <2–70% from 1971 to 2005. A population model incorporating egg predation and herring fishing explains the major population trends of Georges Bank herring over four decades and predicts that, when the haddock population is high, seemingly conservative levels of fishing can still precipitate a severe decline in the herring population. These findings illustrate how efforts to rebuild fisheries can be undermined by not incorporating ecological interactions into fisheries models and management plans.