It is argued that the considerable resilience of most fish stocks under exploitation is evidence in favour of rather strong density-dependent control of recruitment, even though this may not be apparent in the presence of large stochastic variations. Furthermore, since the density dependent variation required is too large to be produced by variation of fecundity or adult growth, it must derive from variations of survival, and thus occur during the first year of life, which is the only period when mortality is large enough to be modulated by a sufficiently large amount. We conclude that the mechanism for density-dependent control of recruitment, and the large variations which accompany it, must be sought in the first year of life, perhaps during the egg and larval stages. If the growth of larval fish depends on the abundance of food per larva, then their growth rate will be density dependent, and so also will the time taken to grow to any particular size. If their vulnerability to predators depends on their size – and particularly if the onset of metamorphosis depends on size – then the cumulative mortality due to predation before metamorphosis will also be density dependent, even if the predators impose a constant mortality coefficient. This mechanism leads to the numbers of survivors depending on the initial numbers with the functional form of a Beverton-Holt stock-recruitment curve, given a plausible assumption concerning the density dependence of growth rates. The consequences and implications of this result are discussed, and in particular it is stressed that the mechanism involves the intimate interaction of three trophic levels simultaneously. Since the cumulative mortality also depends explicitly on the abundance of food and that of predators, the mechanism provides a consistent explanation for the simultaneous occurrence during the first year of life of (a) powerful density-dependent control of recruitment, and (b) high stochastic variability caused by external factors.