Although studies have shown influenza vaccines to be effective in preventing death in the elderly population, these findings may be the result of selection bias. We examined the relationship between vaccination, age, underlying morbidity, and probability of death in the upcoming year. Vaccination coverage varied in a curvilinear fashion with age, morbidity, and risk of death. Forgoing vaccination predicted death in those who had received vaccinations in the previous 5 years, but it predicted survival in patients who had never before received a vaccination. We conclude that bias is inherent in studies of influenza vaccination and death among elderly patients.