One of the objectives of knee replacement is to correct flexion deformity, the frequent consequence of rheumatoid arthritis and osteoarthritis. A review of 697 primary and revision replacements carried out between 1969 and 1985 and followed up from 1 to 16 years found that such deformity was present in 61% of knees before the primary operation. Replacement reduced this to 17% and the improvement was usually maintained. The deformity was present in only 21% of the replacements which required revision and the second operation reduced this to 8%. Flexion contractures affected rheumatoid knees more often and more seriously than osteoarthritic knees, but arthroplasty was more successful in correcting the deformity in the former. All of the 11 types of prosthesis used achieved some degree of correction, but the Walldius hinge and the variants of the Freeman condylar design were the most successful. Surprisingly, the best outcome, in terms of pain and reduced need for revision, was found in the rheumatoid knees most seriously deformed before operation, but this association was absent in the osteoarthritic knees. Postoperative deformity in knees without pain or extreme weakness did not appear to influence the patients' ability to walk or to use stairs or a chair, as measured by unexacting tests in the clinic.