In this study, we quantify the reduction in total sigma (sigma(total)) of the Turkish ground-motion predictive models by partially removing the ergodic assumption. By partitioning residuals into four components: repeatable source-specific and site-specific deviations from the population mean as well as site-corrected within-event and source-corrected between-event parts, event-corrected single-station standard deviation (phi(ss)) and source-corrected between-event standard deviation (tau(ss)) models were derived. With site correction, sigma(total) was shown to reduce by 15%. With source correction as well, sigma(total) was shown to reduce by a further 15% at short-and long-period ranges. The developed aleatory variability models are recommended to be used in hazard assessment studies of single sites under the dominant effect of the North Anatolian fault zone. Last but not least, we quantify the total ground-motion variability by computing the model-to-model variability and statistical-error components of epistemic uncertainty corresponding to two recent probabilistic seismic hazard assessment efforts for Turkey.