We carried out a prospective study of 93 patients undergoing surgery for conditions of the rotator cuff during 1994 and 1995. They were assessed before operation and after six months, and four years, using the patient-based Oxford Shoulder Score (OSS), the SF-36 questionnaire and the Constant shoulder score. The response rates were higher for the OSS and SF-36. The correlation coefficients were high (r > 0.5) between all scores at each stage of the study. While all scores improved substantially at six months, the Constant score was reduced significantly at four years. This did not correlate with the patients’ judgement of the change in symptoms or of the success of the operation. Our study suggests that patient-based measures of pain and function can reliably assess outcomes in the medium term after surgery to the shoulder.