OBJECTIVE: To evaluate the effects of case mix, practice patterns, features of the payment system, and facility characteristics on the cost of dialysis. DATA SOURCES/STUDY SETTING: The nationally representative sample of dialysis units in the 1991 U.S. Renal Data System's Case Mix Adequacy (CMA) Study. The CMA data were merged with data from Medicare Cost Reports, HCFA facility surveys, and HCFA's end-stage renal disease patient registry. STUDY DESIGN: We estimated a statistical cost function to examine the determinants of costs at the dialysis unit level. PRINCIPAL FINDINGS: The relationship between case mix and costs was generally weak. However, dialysis practices (type of dialysis membrane, membrane reuse policy, and treatment duration) did have a significant effect on costs. Further, facilities whose payment was constrained by HCFA's ceiling on the adjustment for area wage rates incurred higher costs than unconstrained facilities. The costs of hospital-based units were considerably higher than those of freestanding units. Among chain units, only members of one of the largest national chains exhibited significant cost savings relative to independent facilities. CONCLUSIONS: Little evidence showed that adjusting dialysis payment to account for differences in case mix across facilities would be necessary to ensure access to care for high-cost patients or to reimburse facilities equitably for their costs. However, current efforts to increase dose of dialysis may require higher payments. Longer treatments appear to be the most economical method of increasing the dose of dialysis. Switching to more expensive types of dialysis membranes was a more costly means of increasing dose and hence must be justified by benefits beyond those of higher dose. Reusing membranes saved money, but the savings were insufficient to offset the costs associated with using more expensive membranes. Most, but not all, of the higher costs observed in hospital-based units appear to reflect overhead cost allocation rather than a difference in real resources devoted to treatment. The economies experienced by the largest chains may provide an explanation for their recent growth in market share. The heterogeneity of results by chain size implies that characterizing units using a simple chain status indicator variable is inadequate. Cost differences by facility type and the effects of the ongoing growth of large chains are worthy of continued monitoring to inform both payment policy and antitrust enforcement.