Abstract Background Elderly patients with breast cancer are less likely to be offered surgery, partly owing to co-morbidities and reduced functional ability. However, there is little consensus on how best to assess surgical risk in this patient group. Methods The ability of pretreatment health measures to predict complications was investigated in a prospective cohort study of a consecutive series of women aged at least 70 years undergoing surgery for operable (stage I–IIIa) breast cancer at 22 English breast units between 2010 and 2013. Data on treatment, surgical complications, health measures and tumour characteristics were collected by case-note review and/or patient interview. Outcome measures were all complications and serious complications within 30 days of surgery. Results The study included 664 women. One or more complications were experienced by 41·0 per cent of the patients, predominantly seroma or primary/minor infections. Complications were serious in 6·5 per cent. More extensive surgery predicted a higher number of complications, but not serious complications. Older age did not predict complications. Several health measures were associated with complications in univariable analysis, and were included in multivariable analyses, adjusting for type/extent of surgery and tumour characteristics. In the final models, pain predicted a higher count of complications (incidence rate ratio 1·01, 95 per cent c.i. 1·00 to 1·01; P = 0·004). Fatigue (odds ratio (OR) 1·02, 95 per cent c.i. 1·01 to 1·03; P = 0·004), low platelet count (OR 4·19, 1·03 to 17·12: P = 0·046) and pulse rate (OR 0·96, 0·93 to 0·99; P = 0·010) predicted serious complications. Conclusion The risk of serious complications from breast surgery is low for older patients. Surgical decisions should be based on patient fitness rather than age. Health measures that predict surgical risk were identified in multivariable models, but the effects were weak, with 95 per cent c.i. close to unity.