Recent meta-analyses of studies of the risk of breast cancer associated with hormone replacement therapy agree that little risk is associated with ever-use or short-term use of estrogen replacement. These analyses disagree, however, about the effect of long-duration estrogen use. To understand differences in the findings among the meta-analyses of the effect of long-term use, we investigated the source of heterogeneity among the included studies. We analyzed subgroups by source of controls (community vs hospital), study design (case-control vs follow-up), and types of estrogen. We also examined the effect of modeling assumptions: that before women began estrogen use, those who chose to use estrogen replacement (1) were, or (2) were not, at substantially different risk from those who chose not to use estrogen. We found a small increase in risk in all subgroups of studies except those that used hospital controls. From a homogeneous group of case-control studies using community controls that analyzed the effect of conjugated equine estrogens, we estimated that the risk of breast cancer after 10 years of estrogen use increased by at least 15% and up to 29%.