In order to quantify the effects of apolipoprotein (apo) E on early- and late-onset, sporadic and familial Alzheimer's disease (AD) more precisely we performed a meta-analysis of 42 case-control series published before July 1996. We excluded studies of late-onset AD with fewer than 50 cases and 50 controls. The estimated odds ratio of AD over the age of 65 years associated with the apo epsilon4 allele, relative to apo epsilon3, was 3.18 (95% CI 2.93-3.45, p < 0. 00001). Apo epsilon4 was associated with a significantly higher relative risk of early-onset disease (<65 years) unselected for family history (OR 4.86, 95% CI 3.61-6.54, p < 0.0001), but its impact on early-onset familial disease was weaker (OR 1.48, 95% CI 1. 02-2.16). The estimated odds ratio associated with the epsilon4/epsilon4 genotype, relative to the epsilon3/epsilon3 genotype, was 11.57 (95% CI 8.67-15.44) for all cases over 65 and 61. 44 (95% CI 13.47-280.3) for early-onset disease unselected for family history. Apo epsilon2 was associated with a significantly reduced odds ratio for AD above the age of 65 (OR 0.68, 95% CI 0. 58-0.80, p < 0.00001), but not in familial early-onset AD, where apo epsilon2 may be associated with an increased risk (OR 1.70, 95% CI 1. 02-2.84). We estimate that 60% (95% CI 48-68%) of AD cases over the age of 65 years and 92% of cases below the age of 65 would be attributable to apo E and that apo E probably accounts for less than 50% of the familial aggregation of the disease.