First-principles kinetic Monte Carlo models allow for the modeling of catalytic surfaces with predictive quality. This comes at the price of non-negligible errors induced by the underlying approximate density functional calculation. On the example of CO oxidation on RuO2(110), we demonstrate a novel, efficient approach to global sensitivity analysis, with which we address the error propagation in these multiscale models. We find, that we can still derive the most important atomistic factors for reactivity, albeit the errors in the simulation results are sizable. The presented approach might also be applied in the hierarchical model construction or computational catalyst screening. (C) 2017 Elsevier B.V. All rights reserved.