Epistemic uncertainty in ground motion prediction relations is recognized as an important factor to be considered in probabilistic seismic hazard analysis (PSHA), together with the aleatory variability that is incorporated directly into the hazard calculations through integration across the log-normal scatter in the ground motion relations. The epistemic uncertainty, which is revealed by the differences in median values of ground motion parameters obtained from relations derived for different regions, is accounted for by the inclusion of two or more ground motion prediction relations in a logic-tree formalism. The sensitivity of the hazard results to the relative weights assigned to the branches of the logic-tree, is explored through hazard analyses for two sites in Europe, in areas of high and moderate seismicity, respectively. The analyses reveal a strong influence of the ground motion models on the results of PSHA, particularly for low annual exceedance frequencies (long return periods) and higher confidence levels. The results also show, however, that as soon as four or more relations are included in the logic-tree, the relative weights, unless strongly biased towards one or two relations, do not significantly affect the hazard. The selection of appropriate prediction relations to include in the analysis, therefore, has a greater impact than the expert judgment applied in assigning relative weights to the branches of the logic-tree.