Over 3 years, 32,444 age-0 group Atlantic salmon (Salmo salar) parr tagged with passive integrated transponder tags throughout the River Frome catchment were assigned to one of three groups, nonmigrants, autumn migrants and spring smolts, depending on the detection and the timing of detection at downstream tag readers (situated 8.6 km above the tidal limit). We examined the effect of density at the time of tagging (n·m−2), distance upstream from the tidal limit (km), fish length (mm), Fulton condition index, habitat type (divided into two types, main river and carrier), days after 1 September that each fish was tagged and year (replication) on the proportions of fish in each of the migration groups. Distance upstream from tidal limit was strongly negatively related to the proportion of autumn migrants and positively related to the proportion of spring smolts. Nonmigrants had a lower average body size than migrants, although there were no differences in the sizes of autumn migrants and spring smolts in September prior to migration. Fish density had no effect on migration strategy. A lower proportion of fish migrated as autumn migrants from the smaller carrier habitats than the main river channel. There is some evidence that those parr destined to become autumn migrants underwent a lower mortality rate during September before tagging than those destined to become spring migrants indicating possible physiological or behavioural differences between these two groups of fish at that time. More research into the factors responsible for initiating the autumn migration is required.