A total of 10,474 unicompartmental knee arthroplasties was performed for medial osteoarthritis in Sweden between 1986 and 1995. We sought to establish whether the number of operations performed in an orthopaedic unit affected the incidence of revision. Three different implants were analysed: one with a high revision rate, known to have unfavourable mechanical and design properties; a prosthesis which is technically demanding with a known increased rate of revision; and the most commonly used unicompartmental device. Most of the units performed relatively few unicompartmental knee arthroplasties per year and there was an association between the mean number carried out and the risk of later revision. The effect of the mean number of operations per year on the risk of revision varied. The technically demanding implant was most affected, that most commonly used less so, and the outcome of the unfavourable design was not influenced by the number of operations performed. For unicompartmental arthroplasty, the long-term results are related to the number performed by the unit, probably expressing the standards of management in selecting the patients and performing the operation.