Virtual population analysis requires information on the fishing mortality rate (orabundance) for one age group from each cohort. In many cases available data areinsufficient to estimate these rates for every age group and structural assumptions mustbe imposed to reduce the number of effective parameters. Past assessments of westernAtlantic bluefin tuna (Thunnus thynnus) have reduced the number ofparameters by assuming pre-specified values for the ratios of the fishing mortality rateson adjacent age groups. More recent bluefin tuna assessments have estimated terminalfishing mortality rates (F) on each age with a constraint that restrictschange from one year to the next. We explore the implications of these methods ofestimating terminal-year F through retrospective analyses of the 2006bluefin tuna assessment and through stochastic simulations. The use of pre-specifiedratios for F created strong retrospective biases and may have led tooverly optimistic projections. Constraining annual changes in the terminal-yearF appeared to mute retrospective patterns and resulted in abundanceprojections less prone to spurious initial leaps. Simulation results indicate that theconstraint improves estimation, particularly with moderate to low interannual changes inselectivity.