The prognostic importance of HER-2 status in breast cancer has been investigated extensively, but findings have not been uniform across immunohistochemical studies using fixed, paraffin-embedded specimens. We speculate that studies with an overrepresentation of large tumors might not produce evidence for an independent effect of a single marker because breast tumors of larger size tend to exhibit multiple adverse attributes as the malignancy advances through the metastatic cascade. Further, it has been posited that results from certain studies of biologic markers might be generalizable only to larger tumors because tumor repositories tend to house a disproportionate number of larger tumors. To test our hypothesis that the prognostic effect of HER-2 status might be modified by the size of the tumor, we conducted a survival analysis of a nested case-case sample of 156 women diagnosed with primary breast cancer from 1983 to 1995. Relative risks (RRs) and confidence intervals (CIs) for recurrence in relation to HER-2 status were estimated using a multivariate Cox proportional hazards model. Immunohistochemistry of archival tissue was used to detect HER-2 expression. Positive HER-2 status was associated with recurrence (RR = 4.24, 95% CI 1.30-13.78) among patients with axillary lymph node-positive involvement. This analysis identified an interaction (p < 0.01) between tumor size and overexpression. Stratification by tumor size revealed an increased risk of recurrence associated with HER-2-positive tumors that were <or=2 cm in size (RR = 13.68, 95% CI 1.06-175.76), whereas no such association was observed among patients with HER-2-positive tumors larger than 2 cm. Eligible patients without available tumor tissue tended to have smaller tumors compared to study participants. Our results might partly explain the variation in evidence across HER-2 prognostic studies using fixed tissue and might apply to other putative markers of prognosis in breast cancer.