The Swedish Knee Arthroplasty Register is dependent on the participating clinics regarding accuracy of information. As the register is prospective, and since revision is used as the endpoint in survival analyses, it is important that all revisions of registered primary arthroplasties are reported. To validate the register, we sent a questionnaire to all living patients with 30,796 knees registered as having been operated on from 1975-1995. Of living patients, 99% could be located and 93% answered. We found that one fifth of the revisions had not been reported and that relatively fewer revisions were lost to follow-up during the first decade of the register than in the following years. To investigate whether the Patient Administrative System (PAS), a database based on ICD coding and run by the Swedish health authorities, could be used to locate missing revisions found by the postal survey, we compared this database with the Swedish Knee Arthroplasty Register. 84% of the missing revisions revealed by the postal survey were found by using this method. Hence after the survey and the use of the PAS to find unreported revisions in deceased and non-responding patients, we estimate that 94% of all revisions are accounted for. Apart from a generally higher cumulative revision rate, conclusions reported from the Register in recent years regarding survivorship seem to be unaffected by the underreporting.