The significant number of seismic ground motion records to be considered when designing or assessing civil structures is a common restriction for employing advanced nonlinear dynamic methods. This is because the large computational time involved in the calculation of the nonlinear dynamic response of complex multi-degree-of-freedom systems. There are several strategies to overcome this limitation; however, the reliability estimation of the analyzed systems can be compromised. This research is focused on developing a methodology to achieve a probabilistic and reliable estimation of the seismic response of buildings by tackling the computational effort. To do so, a set of transfer functions extracted from the dynamic response of three building models have been obtained. Then, an optimal transfer function per building is identified as the one maximizing the prediction of engineering demand parameters (EDPs), when each structure is subjected to a large set of ground motions records. Results show that the response of a reduced number of records allows developing an enhanced strategy to obtain reliable results in terms of the main statistical moments of the EDPs. This increased capacity to analyze complex systems in an affordable time has important consequences in the identification of optimal designs in terms of the material-performance ratio, as well as in the estimation of expected seismic risk.