A general framework is presented to develop, test and integrate component models of the distribution and dynamics of Antarctic krill population at various spatial and temporal scales. We suggest that models of increasing complexity be developed iteratively for variability and patchiness of krill abundance. Incremental models should then be compared to statistical descriptions of the observed distribution patterns at various scales of observation to ascertain the plausibility of the model and identify critical processes to be added. An analysis of spatial distribution of krill in the Bransfield Strait area reveals that purely physical models of turbulent redistribution are not sufficient to explain krill distribution at small scales. We therefore propose to develop a modified diffusion-reaction model incorporating spatially variable growth rates of krill, krill loss rates due to predators, and density-dependent attraction of krill to account for the small-scale aggregations.