Marker-loss is a common feature of mark–recapture studies and important as it may bias parameter estimation. A slight alteration in tag-site of double tagged southern elephant seals (Mirounga leonina), marked at Marion Island from 1983 to 2005 in an ongoing mark–recapture program, had important consequences for tag-loss. We calculated age-specific tag-retention rates and cumulative tag-retention probabilities using a maximum likelihood model selection approach in the software application TAG_LOSS 3.2.0. Under the tag-loss independence assumption, double tag-loss of inner interdigital webbing tags (IIT; 17 cohorts) remained below 1% in the first 5 yr and increased monotonically as seals aged, with higher tag-loss in males. Lifetime cumulative IIT tag-loss was 11.9% for females and 18.4% for males, and equivalent for all cohorts. Changing the tag-site to the outer interdigital webbing (OIT; 6 cohorts) resulted in increased and cohort-dependent tag-loss, although the variation (mean ± 95% CI) in cumulative tag-loss probabilities never exceeded 5.3% between cohorts at similar age. Although different studies may homogenize techniques, we advocate the importance of data set-specific assessment of tag-loss rates to ensure greatest confidence in population parameters obtained from mark–recapture experiments. Permanent marking should be implemented where feasible.