In breast cancer, the prognostic impact of COX2 expression varies widely between studies. We examined the prognostic value of COX2 expression in a large cohort of breast cancer patients treated with primary surgery between 1985 and 1994 and explained the variable results of COX2 expression found in the literature. A tissue microarray was constructed of available tumour material, and ER, PgR, HER2, Ki67 and COX2 were examined by immunohistochemistry. Median follow-up was 19 years. Fifty-five percent (n = 369/677) of patients received no systemic treatment. COX2 was scored using a weighted histoscore. Analysis of COX2 expression in two groups based on the median (148; below vs. above) showed an increased hazard ratio (HR) of 1.35 (95% CI 1.05-1.75, P = 0.021) for disease-free survival (DFS) and of 1.39 (95% CI 1.03-1.82, P = 0.016) for overall survival (OS). However, COX2 did not remain independent in multivariate analysis. In patients with hormone receptor positive tumours, COX2 expression had a negative influence on outcome (low vs. high: DFS: HR 1.37, 95% CI 1.07-1.76, P = 0.013). This effect disappeared when endocrine therapy was administered (low vs. high: DFS: HR 0.93, 95% CI 0.51-1.70, P = 0.811) while it remained statistically significant when endocrine therapy was omitted (low vs. high: DFS: HR 1.48, 95% CI 1.12-1.94, P = 0.005). Our results show that COX2 plays a role in hormonal pathways. Our results can explain the results found in previously published studies.