Multisite procedure-based randomized trials may be confounded by performance variability and variability among sites. Therefore, we studied variability in mortality and stroke after patients were randomized to surgical (SAVR) or transcatheter aortic valve replacement (TAVR) in the Placement of Aortic Transcatheter Valves-2A (PARTNER-2A) randomized trial.Patients at intermediate risk for SAVR were randomized to SAVR (n = 1017) or TAVR (n = 1011) with a SAPIEN XT device (Edwards Lifesciences, Irvine, Calif) at 54 sites. Patients were followed to 2 years. A mixed-effect model quantified variability at intersite and intrasite levels.There were 336 deaths (SAVR 170, TAVR 166) and 176 strokes (SAVR 85, TAVR 91). Intersite variability for mortality was similar across sites for SAVR (hazard ratios ranging from 0.52-1.93 among sites) and TAVR (hazard ratios ranging from 0.49-2.03), but intersite variability for stroke was greater for SAVR (hazard ratios ranging from 0.44-2.26) than for TAVR (no detectable variability). Case mix and lower site trial volume accounted for 37% of mortality intersite variability for SAVR and 73% for TAVR, but only 14% for stroke for SAVR. Intrasite mortality hazard ratios demonstrated all but 1 site's 95% confidence interval overlapped 1.0, indicating generally similar SAVR and TAVR mortalities within sites.Intersite variability was similar for mortality in SAVR and TAVR, but variability for stroke was greater for SAVR than for TAVR. Intrasite events were similar for both SAVR and TAVR. These findings suggest that in performance-based trials, site variability and its sources should be taken into account in analyzing and interpreting trial results.