This dissertation examines how the selection and scaling of natural earthquake records influences the variability in estimated seismic demands on inelastic reinforced concrete structures during nonlinear time-history analysis. The author uses 148 recorded ground motions matched to the EC8 target spectrum to generate sample sets of 4, 7, and 11 records with best, average, and worst matches. Records are scaled based on spectral intensity and seismicity levels, requiring 396 analyses. Results are assessed in terms of mean and peak displacement ductility demands on a 3-story structure's critical floors under 2214 earthquake cases.