BACKGROUND--The association between coffee drinking and risk of coronary heart disease remains controversial despite many epidemiological studies. A meta-analysis was carried out on these studies to resolve some of the uncertainties. Particular attention was paid to details of study design. METHODS--Eight case-control studies and 15 cohort studies were analysed. Weighted, fixed effects linear regression of log relative risks (or odds ratios) was used to pool the study results. The pooling procedures were performed separately by study design, sex, coronary heart disease end points, smoking habit, and period of study. RESULTS--The pooled case-control odds ratio (for the effect of drinking five cups of coffee/day v none) was 1.63 (95% confidence interval (95% CI) 1.50 to 1.78). The pooled cohort study relative risk (five cups/day v none) was 1.05 (95% CI 0.99 to 1.12). The discrepancy between the pooled case-control and cohort study results could not be attributed to differences in the end points chosen, period of study, or to confounding by smoking status or sex. CONCLUSIONS--The cohort study data suggest very little excess risk of coronary heart disease among habitual coffee drinkers. The case-control data do not rule out an increased risk of heart disease among a subgroup of people who acutely increase their coffee intake. Further epidemiological studies are needed to assess the risk of drinking boiled or decaffeinated coffee.