Repeated risk factor measurements reduce the amount of random error due to intraindividual variation. Annual measurements of serum cholesterol and systolic blood pressure were done in a cohort of 878 men, aged 40-59 years, between 1960 and 1970. This study examined for both risk factors the association with the incidence of myocardial infarction (MI) during follow-up till 1985. Mean values from repeated measurements yielded stronger statistical associations with MI incidence than single observations. However, when measurements were extended over too many years in this middle-aged cohort, a bias occurred due to selective loss of high-risk individuals as early cases. This was especially the case for serum cholesterol. Taking account of trends in regression analysis did not show any advantage over the use of mean values.