Our objective was to examine the rate of revision and its predictive factors in patients undergoing total shoulder arthroplasty (TSA). We used prospectively collected data from the Mayo Clinic Total Joint Registry to examine five-, ten- and 20-year revision-free survival following TSA and the predictive factors. We examined patient characteristics (age, gender, body mass index, comorbidity), implant fixation (cemented versus uncemented), American Society of Anesthesiologists class and underlying diagnosis. Univariate and multivariable adjusted hazard rates were calculated using Cox regression analysis. A total of 2207 patients underwent 2588 TSAs. Their mean age was 65.0 years (19 to 91) and 1163 (53%) were women; osteoarthritis was the underlying diagnosis in 1640 shoulders (63%). In all, 212 TSAs (8.2%) were revised during the follow-up period. At five, ten and 20 years, survival rates were 94.2% (95% confidence interval (CI) 93.2 to 95.3), 90.2% (95% CI 88.7 to 91.7) and 81.4% (95% CI 78.4 to 84.5), respectively. In multivariable analyses men had a higher hazard ratio of revision of 1.72 (95% CI 1.28 to 2.31) (p < 0.01) compared with women, and those with rotator cuff disease had a hazard ratio of 4.71 (95% CI 2.09 to 10.59) (p < 0.001) compared with patients with rheumatoid arthritis. We concluded that male gender and rotator cuff disease are independent risk factors for revision after TSA. Future studies are needed to understand the biological rationale for these differences. © 2011 British Editorial Society of Bone and Joint Surgery.