Central to understanding krill population dynamics is knowledge of their population structure. To examine this we used length-frequency distributions from 142 weeks of sampling (n= 23996 krill) of 3 predator species breeding at South Georgia and 12 weeks of sampling (n = 10 252 krill) from scientific nets from the same area over the summers of 1991-1997. In comparing the 5 years with both predator and net samples, despite differing selectivities and spatio-temporal circumscriptions, both were sampling the same overall krill population. Greatest similarity results from comparing net samples with samples from Antarctic fur seal and macaroni penguin combined; least temporal variation occurs in predator samples from late summer (March). From the 7-year predator time-series, within-year variation was greatest in 1991 and 1994, both years of low krill biomass at South Georgia. In both of these years large krill dominated during December but were completely replaced by small krill by February. The mean length of the March krill population showed a regular increase from 1991 to 1993, fell to a minimum in 1994 and thereafter increased steadily to 1997. Using these data in conjunction with putative size/age-group cohorts in the length-frequency distribution, we suggest that year's of high mean krill length reflect failure of small krill to recruit into the population, producing a period of low krill biomass in the following year. Similar recruitment failure in the same years is evident in krill populations in the Antarctic Peninsula region to the south, indicating large-scale events. This supports suggestions of periodic fluctuations in krill production and recruitment which may relate directly to physical phenomena such as cycles in the distribution and extent of sea-ice.