The T4336C mitochondrial genetic variant was associated with Alzheimer disease in several previous studies. Recent investigations, however, failed to confirm this association. We tested this association in newly diagnosed Alzheimer disease cases and controls of similar age and gender recruited from an established HMO serving Seattle, Washington and surrounding areas. In this, the largest case-control study reported to date, the T4336C variant was not associated with Alzheimer disease overall (present in 6 of 236 cases and 7 of 328 controls; odds ratio = 1.20, 95% CI 0.33 to 4.22). There was evidence of effect modification by Apolipoprotein E (APOE) status--among subjects with an APOE epsilon 4 allele, the T4336C variant was associated with disease (present in 5 of 139 cases and none of 82 controls; odds ratio = infinity, 95% CI 0.73 to infinity). APOE may be an important modifier of the T4336C effect, potentially explaining variable findings across previous studies. Alternatively, the positive findings reported to date may simply reflect the problem of "type I" error inherent in genetic association studies. Substantially larger samples than are currently available would be required to resolve this question. G5460(A/T) variants were also investigated and found not to be associated with Alzheimer disease.