Abstract Background Primary endocrine therapy is used as an alternative to surgery in up to 40 per cent of women with early breast cancer aged over 70 years in the UK. This study investigated the impact of surgery versus primary endocrine therapy on breast cancer-specific survival (BCSS) in older women. Methods Cancer registration data for 2002–2010 were obtained from two English regions. A retrospective analysis was performed for women with oestrogen receptor (ER)-positive disease, using statistical modelling to show the effect of treatment (surgery or primary endocrine therapy) and age and health status on BCSS. Missing data were handled using multiple imputation. Results Cancer registration data on 23 961 women were retrieved. After data preprocessing, 18 730 of 23 849 women (78·5 per cent) were identified as having ER-positive disease; of these, 10 087 (53·9 per cent) had surgery and 8643 (46·1 per cent) had primary endocrine therapy. BCSS was worse in the primary endocrine therapy group than in the surgical group (5-year BCSS rate 69·4 and 89·9 per cent respectively). This was true for all strata considered, although the difference was less in the cohort with the greatest degree of co-morbidity. For older, frailer patients the hazard of breast cancer death had less relative impact on overall survival. Conclusion BCSS in older women with ER-positive disease is worse if surgery is omitted. This treatment choice may contribute to inferior cancer outcomes. Selection for surgery on the basis of predicted life expectancy may permit choice of women for whom surgery confers little benefit.