The data from tag-recapture experiments, which are used to help understand animal behaviour and dynamics, and to provide input data for population models such as stock assessments, are affected by mortality associated with tagging and by tag shedding. These processes introduce bias and uncertainty into parameters estimated in population models such as tuna stock assessments. The causes and magnitudes of tag shedding and post-release mortality in tuna tagging experiments are not well understood. We analysed data from tuna tagging experiments in the Western Pacific (330,000 releases) and Indian Oceans (168,000 releases) to investigate factors affecting post-release mortality and tag shedding. Tag return rates were modelled as functions of the tagger identity, tagger experience, tagging assistant, tagging station, treatment of the fish, use of oxytetracycline, tuna species, and size at release. The release event was included in models as a fixed effect, so that differences in recapture rate among release events did not affect other parameter estimates. We found differences in tag return rates among taggers and tagging assistants, with tagger experience, and between tagging stations. Substantially lower return rates were associated with some types of damage to fish and with internally implanted tags, and when oxytetracycline was injected. Return rates varied with tuna size and species. In the Western Pacific yellowfin and bigeye return rates were more affected by some covariates than were skipjack, while differences were not observed in the Indian Ocean, where sample sizes were smaller. Results suggest that tagging mortality may be quite high, and that more care and better recording of fish treatment would increase the reliability of assessment inputs. We provide new effective release numbers that have been adjusted to allow for estimated tagging mortality and tag shedding.