Background: We propose a model to characterize the variation in total knee arthroplasty (TKA) episode payments in the U.S. Medicare population to establish a baseline prior to the full implementation of the Comprehensive Care for Joint Replacement (CJR) model. Methods: We identified TKA episodes in Medicare Part A (100% sample) from 2014 to 2016 (n = 717,690) and compared 90-day episode payments across years and geographic regions. We fit hierarchical models that regressed episode payments on patient-level fixed effects (age, sex, race, comorbidities) and region-level (U.S. Census Regions) and hospital-level random effects. Random-effect estimates were used to characterize risk-adjusted hospital cost performance. We ranked hospitals (n = 3,217) in each region by their cost performance estimate and constructed 95% confidence intervals to visualize high and low-performing hospitals. Results: During this period, the mean Part A episode payments declined throughout the United States ($18,665 to $16,978; p < 0.001), primarily because of decreased post-acute care payments ($6,401 to $4,873; p < 0.0001). The 90-day readmission rates fell by nearly 20% (7.2% to 5.8%; p < 0.001). We found significant variation (p < 0.05) in risk-adjusted episode payments, post-acute care utilization, and readmission rates across regions and even hospitals. The share of hospitals in each geographic region that were low-performance outliers for episode payments ranged from 13% to 31% and those that were high-performance outliers ranged from 16% to 30%. Conclusions: Medicare Part A payments for TKA episodes were decreasing prior to the CJR model because of decreases in both post-acute care utilization and hospital readmissions. A significant variation in risk-adjusted hospital cost performance remained. Our results provide a baseline against which to measure the impact of alternative payment models and a methodology by which to measure hospital-level performance, which can be compared with peer hospitals and national benchmarks.