Survival and reproductive rate estimation requires following uniquely identified individuals through time, and many statistical models assume markings used to identify individuals are permanent. However, survival rates are underestimated when single marks are lost, since the models will effectively score those animals as dead. In order to account for mark loss, some researchers use a double‐mark approach, assuming the probability of losing one mark is independent of losing the other one. Therefore, mark loss can be estimated using animals that have lost one mark. Using a 17‐year dataset of southern elephant seals (Mirounga leonina) marked with permanent brands and two cattle tags in their hind flippers, we were able to compare tag loss and survival rate estimates with and without the assumption of independent tag loss with respect to age, sex, and wean mass. We demonstrate the assumption of independent tag loss is not valid, showing it is more likely for an animal to lose both tags than just one or the other. The assumption of independent tag loss leads to an underestimate of survival rates which in turn leads to underestimates of population growth rate. In addition, tag loss rates are different by sex and age, with older males more likely to lose tags. Tag loss is also a quadratic function of wean mass through age two, with smaller and larger animals more likely to lose both tags. Such differences are possibly due to differences in behavior, flipper growth, and immune response. Using a Bayesian approach, we will be able to use our tag loss estimates as priors in future analyses for a subset of marked animals that only have flipper tags. With this population, the independent tag loss models are more likely to incorrectly estimate a declining population (growth rate < 1.0), potentially leading to unwarranted management action. To account for non‐independent mark loss in survival rate studies, we recommend researchers use at least two forms of marking on at least a subset of animals. However, neither form of marking need be permanent as long as mark loss is independent between the different forms.