With the advent of the time domain analyses, preferred over the design methods based on the response spectrum idea, the use of time histories in the seismic analysis of dams is common practice. Selection of the ground motions to be used in time history analysis is a crucial task as the seismic demands show large variance depending on the chosen records. In order to reduce this variability and predict the true' demand related to the seismic hazard conditions of the site, the selected ground motions are usually scaled and/or modified. Commonly utilised ground motion scaling techniques (i.e. scaling to the target spectrum, maximum incremental velocity and spectrum matching) are evaluated in this study for determining the efficiency and accuracy of the scaling technique in predicting the target demands on the concrete gravity dams. Different canyon geometries and moduli ratios were considered in a robust soil-structure-reservoir interaction framework including the ground motion variability through the use of a large ensemble of motions. The minimum number of required motions for consistent estimation of seismic demands on gravity dams was determined to be higher than current practice. Demand parameters corresponding to the non-linear response were determined to be underestimated by the scaled sets.