OBJECTIVE. The aim of this study is to describe the practice variation of knee replacements (KRs) in the elderly ( > or = 65) over time from 1985-1990 in terms of the number of primary, bilateral, and revision KRs; the extent of large area variation in performance rates; and the degree to which demographic variables are the determinants of area rates. DATA SOURCES/STUDY SETTING. Data analyzed are from every hospital in the United States that performed a KR on a Medicare patient during the study period. Data were obtained from the MEDPAR, HISKEW, and denominator files of the Medicare Statistical System. STUDY DESIGN. This is a cohort study of all Medicare beneficiaries who received a KR between 1985 and 1990. The dependent variable in the analyses was the count of the KRs performed in each area. DATA COLLECTION/EXTRACTION METHODS. This is a population-based sample of Medicare enrollees in the United States. All hospitalizations for Medicare-reimbursed KRs were included in the initial data set. Exclusion criteria were used to identify the Medicare covered population with a definite KR. These criteria resulted in 7.3 percent exclusions and a final set of 414,079 KR hospitalizations. PRINCIPAL FINDINGS. The number of Medicare-funded KRs increased in each of the study years corresponding to an annual rate of increase of 18.45 percent. The likelihood of receiving a KR was a function of age, gender, and race. For each year, KRs were almost-twice as likely to be performed on women than on men. The odds of whites getting the surgery were over 1.5 times greater than for blacks. Even after adjusting for demographic factors, significant regional variation remained. CONCLUSIONS. Much about area variation and the rate of growth in KR rates remains unexplained. For answers to emerge, better data and different types of studies are required.