BACKGROUND: Hip and knee replacement are some of the most frequently performed surgical procedures in the world. Resurfacing of the hip and unicondylar knee replacement are increasingly being used. There is relatively little evidence on their performance. To study performance of joint replacement in England, we investigated revision rates in the first 3 y after hip or knee replacement according to prosthesis type. METHODS AND FINDINGS: We linked records of the National Joint Registry for England and Wales and the Hospital Episode Statistics for patients with a primary hip or knee replacement in the National Health Service in England between April 2003 and September 2006. Hospital Episode Statistics records of succeeding admissions were used to identify revisions for any reason. 76,576 patients with a primary hip replacement and 80,697 with a primary knee replacement were included (51% of all primary hip and knee replacements done in the English National Health Service). In hip patients, 3-y revision rates were 0.9% (95% confidence interval [CI] 0.8%-1.1%) with cemented, 2.0% (1.7%-2.3%) with cementless, 1.5% (1.1%-2.0% CI) with "hybrid" prostheses, and 2.6% (2.1%-3.1%) with hip resurfacing (p < 0.0001). Revision rates after hip resurfacing were increased especially in women. In knee patients, 3-y revision rates were 1.4% (1.2%-1.5% CI) with cemented, 1.5% (1.1%-2.1% CI) with cementless, and 2.8% (1.8%-4.5% CI) with unicondylar prostheses (p < 0.0001). Revision rates after knee replacement strongly decreased with age. INTERPRETATION: Overall, about one in 75 patients needed a revision of their prosthesis within 3 y. On the basis of our data, consideration should be given to using hip resurfacing only in male patients and unicondylar knee replacement only in elderly patients.