Complex analysis methods such as non-linear time-history analyses (NTHA) are often required for the design of non-standard bridges. The selection of the ground motions for the NTHA is a crucial task as the results of the analyses show a wide variability according to the selected records. In order to predict the demand in accordance with the seismic hazard conditions of the site, the selected motions are usually modified by scaling procedures. Within this context, the performance of two scaling methods, namely the Modal Pushover Based Scaling (MPS) and ASCE/SEI procedures, are compared for the NTHA of a large bridge, the Demirtas Viaduct (longitude 29.10 degrees, latitude 40.28 degrees), in this study. The system comprised of 28 spans was idealised with two different analytical models in order to assess the effect of the modelling on the scaling procedures results. The effects of the hazard level on the scaling results were evaluated. The required number of motions for conducting effective analyses, i.e. the minimum number of motions for estimating the target goals, was investigated at different hazard levels. MPS reduced the dispersion considerably more than the ASCE scaling, indicating sets can be formed with fewer motions to predict the target levels.