During three summer surveys at Prince Edward Island (PEI), southern Indian Ocean (2001, 2004 and 2008), 416 southern elephant seals Mirounga leonina were inspected for identification tags. In all, 42 seals that had been tagged as weaned pups at their natal site were found on Marion Island (MI), 38 of which could be individually identified by resighting their tag numbers. The majority of the MI-tagged seals were yearlings or subadults, and all but one were hauled out at PEI for the annual moult. The attendance rate of the known individuals at their natal island during the annual moult was only 40%, based on their resighting histories. This was significantly lower than the 77 ± 6% moult attendance rate estimated for a random MI population sample drawn from the same cohorts (based on 10 000 replications). Annual resight probabilities (considering all haulout phases) was 58% per annum for the MI seals seen at PEI, and 80 ± 4% for the simulation. Seasonal and annual absences of seals from MI violate the ‘homogeneity of capture’ assumption of mark–recapture models. When multiple sightings during any year are treated as a single sighting, resights during other haulouts (e.g. breeding) compensate only partially for absences during the moult. Therefore, mark–recapture studies undertaken in archipelagos should ideally include both marking and resighting of individuals on all islands which will allow discrimination between mortality and local migration.