Temperature control of Atlantic salmon (Salmo salar) smolt migration was tested using a novel technique allowing nearly continuous monitoring of behavior with complete control over environmental conditions. Parr and presmolts were implanted with passive integrated transponder tags, placed in simulated streams, and monitored for upstream and downstream movements. Beginning 18 April, temperature was increased 1??C every third day (advanced), fourth day (ambient), and tenth day (delayed). Smolt downstream movements were initially low, peaked in mid-May, and subsequently declined under all conditions. Parr downstream movements were significantly lower than those of smolts in all treatments (0.8 ?? 0.5 movement??day-1 versus 26.5 ?? 4.5 movements??day-1, mean ?? SE) and showed no increase. At delayed temperatures, smolts sustained downstream movements through July; those under ambient and advanced conditions ceased activity by mid-June. Initiation and termination of downstream movements occurred at significantly different temperatures but at the same number of degree-days in all treatments. Physiological changes associated with smolting (gill Na+,K +-ATPase activity and plasma thyroxine) were coincident with behavioral changes. This is the first evidence of a behavioral component to the smolt window. We found that temperature experience over time is more relevant to initiation and termination of downstream movement than a temperature threshold. ?? 2005 NRC Canada.