Atlantic salmon (Salmo salar) populations have suffered declines across their range in recent decades, largely attributed to decreasing marine survival rates. The first few months at sea are thought to be a time of particular vulnerability, but investigation into drivers of survival is limited by a lack of knowledge of migratory paths. Here, we model the early marine migration of Atlantic salmon from Scottish and Irish rivers over 27 years. Movement is simulated over the first 3 months at sea using a Lagrangian particle tracking model (FVCOM i-state configuration model) coupled with an active swimming model that is dependent on ocean currents, salinity, and compass direction. Our model is driven by the reanalysis of a high-resolution ocean model (Scottish Shelf Waters Reanalysis Service). Differences in the speed of migrations, the proportion of time spent in different oceanographic regions, and the proportion of migrations reaching the Norwegian Sea are seen between years. These differences are related to changes in local ocean conditions: years with lower on-shelf salinity, stronger on-shelf northwest currents, and stronger shelf-edge currents were associated with greater migration success. Within years, differences in modelled migrations between rivers were best explained by their minimum distance from the continental shelf edge.