This study attempts to explain the unexpected finding of an inverse population (ecological) relationship between mean systolic blood pressure levels and stroke death rates in 25 years follow-up of the Seven Countries Study, a cross-cultural study of cardiovascular disease. Sixteen cohorts of all men aged 40-59 in seven countries (one cohort in the USA, two in Finland, one in the Netherlands, three in Italy, two in Croatia (former Yugoslavia), three in Serbia (former Yugoslavia), two in Greece, two in Japan) were surveyed from 1958 to 1964. Risk factors and personal characteristics were measured and follow-up for vital status and cause of death was then carried out over 25 years. Analyses were based on comparisons of mean levels of risk factors and death rates within and among the 16 cohorts. Mean entry population levels of systolic blood pressure among the cohorts were strongly and inversely related with their 25-year stroke death rates (R -0.55; CI -0.81 and -0.06; p = 0.0276). Within cohorts in contrast, the individual relation of blood pressure and stroke was strongly positive and significant in 14 of the 16 cohorts. Mean population levels of serum cholesterol were inversely and strongly related to stroke death rates (R -0.79; CI -0.92 and -0.46; p = 0.0003), while the partial correlation coefficient of systolic blood pressure, computed in models including serum cholesterol, became small and not significant (-0.05; CI -0.55 and +0.48; p = 0.8537). Age at death for stroke (average 68.9 +/- 7.1 years) was significantly higher than age at dath from myocardial infarction and sudden death (average 65.8 +/- 7.8 years) suggesting a competition effect between the conditions. Multivariate models including population average systolic blood pressure and serum cholesterol provided no added explanation for the lack of direct and significant relationship of population blood pressure with stroke death rates. They were based on these variables: age at stroke death, age at myocardial infarction death or and sudden death, death rates from myocardial infarction and sudden death, the interaction term of systolic blood pressure with serum cholesterol and the multivariate coefficients for systolic blood pressure from Cox models run in individuals. Similar findings were obtained using diastolic instead of systolic blood pressure and excluding the Japanese cohorts. The paradox of the inverse ecologic relation of population blood pressure and stroke mortality and a direct relation for individual is only partly explained by the cofounding effect of population mean serum cholesterol levels. An effect of low cholesterol levels on excess stroke mortality cannot be excluded. A major limitation of the study was our inability to segregate thrombotic from heamorrhagic strokes.