The size structure of Arctic charr (Salvelinus alpinus) populations was analysed using published information; 44% of populations were bimodal, with both large (normal) and small (dwarf) morphs occurring within a cohort. The remaining populations were unimodal, consisting of normal-sized or of stunted adults. Bimodal populations increased in frequency with latitude and were characteristic of large, deep lakes with few fish species. The age and size at which bimodality developed, the size difference between the morphs, and the frequency of cannibalism in charr populations increased with latitude. A variety of genotypic and phenotypic explanations are examined. The evidence for specific differences between the morphs is unconvincing. Various competition and predation hypotheses fail to explain the occurrence of bimodality. Cannibalism does not cause bimodality despite being strongly associated with it since bimodality develops before charr become cannibalistic. Much of the variation in charr size structure is suggested to be a consequence of increased seasonality in food supply in more northerly environments, coupled with feeding size thresholds. The latter result in larger members of a cohort being able to maintain growth rates on seasonally abundant prey while smaller individuals which cannot catch these items form a second mode of more slowly growing fish. Bimodality is documented in a number of other, predominantly northern, fish species.