During the winter of 1989-1990, influenza type A(H3N2) circulated widely, causing excess morbidity and mortality nationwide. From November through April, 1989-1990, hospitalized cases of pneumonia and influenza occurring among noninstitutionalized individuals 65 or more years of age were identified by 20 acute care hospitals in southern lower Michigan. These cases were group matched on age, sex, race, and zip code to randomly sampled, community-based controls from a comprehensive listing of Medicare beneficiaries residing in the study area. Self-reported data were collected from cases and controls on influenza vaccine status for the 1989-1990 season and on a number of other factors which could have influenced vaccination status or outcome. Questionnaires were completed by 1,907 individuals, 449 of whom were cases, resulting in an overall response rate of 76%. A community-based influenza surveillance system was implemented to determine the timing and intensity of viral activity and influenza-like illness. Vaccine effectiveness in preventing overall pneumonia and influenza hospitalizations was estimated by logistic regression. During the 3-month period of surveillance-confirmed peak influenza type A(H3N2) circulation, vaccine effectiveness was 45% (95% confidence interval 14-64, p = 0.009). However, during the 3-month period of low or absent virus activity, identical methodology and model specification resulted in an effectiveness estimate of 21% that was not statistically different from zero (p = 0.36). The effectiveness determined during the peak period of virus circulation is felt to be a conservative estimate, since agents other than influenza are responsible for pneumonia and influenza hospitalizations, even during times of peak influenza activity.