Geographic variations in practices and expenditures have been widely documented, leading to concerns that care in some regions is clinically suboptimal and/or economically inefficient. Our objectives are to determine the extent and sources of geographic variation in Medicare expenditures per patient with end-stage renal disease (ESRD) per year. The study population included all patients with ESRD with Medicare as primary payer during 1997 (n = 284,670). Medicare expenditures were summarized at the hospital referral region (HRR) level. Using regression analysis, we estimated the relationship between expenditures and demographics, case mix, dialysis provider characteristics, distribution of patients across renal replacement therapy modalities, standardized hospitalization ratios, and healthcare wages. Spending per patient-year varied threefold across HRRs, ranging from $17,791 to $59,025 (mean, $38,966 +/- $6,774 [SD]). The regression equation explained 80% of this variation. Although several demographic and case-mix indicators that have been related to spending at the individual level were statistically significant predictors of spending at the HRR level, they did not show enough geographic variation to explain a large fraction of spending variation. Rather, patient distributions across renal replacement modalities, hospitalization patterns, and healthcare wages were the most powerful predictors of spending. Compared with Medicare generally, both the mean and SD of ESRD expenditures were approximately seven times larger. The substantial geographic variability in expenditures for patients with ESRD indicates the potential for improving efficiency and quality of care. Interventions designed to increase transplantation rates, ensure access to peritoneal dialysis, and reduce hospitalization appear most promising.