Influenza vaccination policy in most high-income countries attempts to reduce the mortality burden of influenza by targeting people aged at least 65 years for vaccination. However, the effectiveness of this strategy is under debate. Although placebo-controlled randomised trials show influenza vaccine is effective in younger adults, few trials have included elderly people, and especially those aged at least 70 years, the age-group that accounts for three-quarters of all influenza-related deaths. Recent excess mortality studies were unable to confirm a decline in influenza-related mortality since 1980, even as vaccination coverage increased from 15% to 65%. Paradoxically, whereas those studies attribute about 5% of all winter deaths to influenza, many cohort studies report a 50% reduction in the total risk of death in winter--a benefit ten times greater than the estimated influenza mortality burden. New studies, however, have shown substantial unadjusted selection bias in previous cohort studies. We propose an analytical framework for detecting such residual bias. We conclude that frailty selection bias and use of non-specific endpoints such as all-cause mortality have led cohort studies to greatly exaggerate vaccine benefits. The remaining evidence base is currently insufficient to indicate the magnitude of the mortality benefit, if any, that elderly people derive from the vaccination programme.