Data on breeding population size and breeding success in gentoo penguins at Bird Island, South Georgia from 1977-1992 are used, in conjunction with empirical (and some hypothetical) data on survival and recruitment rates, to model the fluctuations in breeding populations, taking account of variations introduced by good and bad years (as classified on the basis of breeding success). There is generally good agreement between observed and predicted breeding populations, except in four years, when major population changes (large decreases followed by substantial increases) occurred. Three of these years were associated with reduced availability of krill, one with very cold winter and spring conditions. Comparing model and reality indicates that deferred breeding could account for the discrepancy in one year, and for part of the differences in two other years, when mortality rates must also have been higher. In the remaining year, when the increase in population greatly exceeded the preceding decrease, it is likely that some immigration occurred. Detailed field studies from 1987-1991 established that the population decline in 1988 was indeed attributable to substantial deferred breeding coupled with higher rates of adult mortality. Emigration was most unlikely to be involved; no data are available on immigration. Gentoo penguin population dynamics are disproportionately effected by the consequences of infrequent bad years; any increase, natural or artificial, in the frequency of such events might have serious consequences for population trends.