Previous assessments of the efficacy of inactivated influenza A vaccines have met with only limited success. Low challenge rates, lack of a stable population and inadequate information on vaccine and control groups have made results difficult to interpret. This work was undertaken to obtain a more accurate assessment of the protective effect of inactivated vaccines and of natural infection in an era of influenza A drift following the emergence of a new pandemic sub-type. A trial of inactivated influenza vaccines was carried out in a boys' boarding school from November 1970 to October 1975. Techniques were developed for the estimation of circulating antibodies to the haemagglutinin and neuraminidase. There was no technique suitable for the estimation of neuraminidase antibodies on a large scale. The mechanism of neuraminidase-haemagglutination-inhibition was investigated and a novel technique was evolved. Estimation of antibodies to both surface antigens made it possible to assess more completely the response to natural infection and vaccination. During the trial period all influenza-like illness was investigated and the incidence of sub-clinical infection with influenza A ascertained. Two major outbreaks of influenza A occurred (in 1972 and 1974) during the study period and one (in 1976) shortly after the end of the trial. Antibody produced after primary natural infection with a new sub-type of influenza A remained detectable for some years as did vaccine-induced antibody. Previous natural infection with influenza A influenced vaccine response and the effect of natural challenge. The antibody response to re-vaccination with the same strain was poor. The protective effect of natural infection and vaccination on challenge with later variants was studied during the outbreaks. Influenza A inactivated vaccines were shown to have a considerable protective effect only when the vaccine strain was an immediate precursor of the challenge strain. Natural infection gave much greater protection against subsequent natural challenge.