Background Patient-level risk factors for delayed graft function (DGF) have been well described. However, the Organ Procurement and Transplantation Network definition of DGF is based on dialysis in the first week, which is subject to center-level practice patterns. It remains unclear if there are center-level differences in DGF and if measurable center characteristics can explain these differences. Methods Using the 2003 to 2012 Scientific Registry of Transplant Recipients data, we developed a hierarchical (multilevel) model to determine the association between center characteristics and DGF incidence after adjusting for known patient risk factors and to quantify residual variability across centers after adjustment for these factors. Results Of 82,143 deceased donor kidney transplant recipients, 27.0% developed DGF, with a range across centers of 3.2% to 63.3%. A center's proportion of preemptive transplants (odds ratio [OR], 0.83; per 5% increment; 95% confidence interval [95% CI], 0.74-;0.93; P = 0.001) and kidneys with longer than 30 hr of cold ischemia time (CIT) (OR, 0.95; per 5% increment; 95% CI, 0.92-;0.98; P = 0.001) were associated with less DGF. A center's proportion of donation after cardiac death donors (OR, 1.12; per 5% increment; 95% CI, 1.03-;1.17; P < 0.001) and imported kidneys (OR, 1.06; per 5% increment; 95% CI, 1.03-;1.10; P < 0.001) were associated with more DGF. After patient-level and center-level adjustments, only 41.8% of centers had DGF incidences consistent with the national median and 28.2% had incidences above the national median. Conclusion Significant heterogeneity in DGF incidences across centers, even after adjusting for patient-level and center-level characteristics, calls into question the generalizability and validity of the current DGF definition. Enhanced understanding of center-level variability and improving the definition of DGF accordingly may improve DGF's utility in clinical care and as a surrogate endpoint in clinical trials.