The efficacy of the influenza vaccine is often understimated, due to the dilution of the outcome by noninfluenzal illnesses. We thus explored the methodology to evaluate the effect of the inactivated influenza vaccine under the following strict conditions: an assessment of the effectiveness on clinical illness among healthy adults in a small-scale mixed epidemic during the 1991-1992 season. The vaccine antigens included were A/Yamagata/32/89 (H1N1), A/Beijing/352/89 (H3N2), and B/Bangkok/163/90. Two indices were analyzed: "vaccine efficacy", a comparison between the vaccinees and the nonvaccinees; and "antibody efficacy", a comparison between those with and those without a protective level of pre-epidemic hemagglutination-inhibition (HAI) antibody. The odds ratio (OR) and its 95% confidence interval (95% CI) was calculated by the logistic regression model. A decrease in the age-adjusted OR of vaccination was not statistically significant: 0.54 (95% CI: 0.19-1.53) corresponding to vaccine efficacy (1-OR) of 46% (-53% to 81%). Among the vaccinees, a significantly decreased OR in those with a higher titer to A/Beijing was observed: 0.14 (0.02-0.92) adjusted for the mutual effects of pre-epidemic antibodies to different vaccine antigens. The adjusted ORs thus calculated for A/Yamagata and B/Bangkok were not found to be statistically significant. The antibody efficacy (1-OR) was estimated to be 86% (8% to 98%) against illnesses related to A/Beijing-like viruses. The product of antibody efficacy (86%) and the proportion of those who achieved a protective level of antibody after vaccination (73% for A/Beijing strain) was 63%, which is theoretically equivalent to the vaccine efficacy. Thus, the antibody efficacy is considered to be an important index, while the vaccine efficacy against clinical illnesses is easily disturbed by extraneous factors in the field trials.