Modern approaches for the assessment of the seismic behavior of structures (i.e. PBEE) are often based on nonlinear dynamic analyses. In this respect, the selection of ground motion records is of paramount importance, because of the record-to-record variability of the structural response. In the past, selection of natural records were often performed based on the compatibility between the response spectra and a target spectrum, usually derived from a PSHA or a code spectrum. Spectral matching can be considered the most commonly proposed earthquake record selection, currently adopted by many seismic codes. The main problem of this approach is that the record-to-record variability can result excessive, obtaining a considerable scatter in the non-linear response. For these reasons, it is usually preferable to synthetize the seismic hazard properties by using a scalar or vector-valued Intensity Measure (IM). Nevertheless, the selection of efficient, sufficient and practical IMs lead to different approaches without converging toward a unified approach. Consequently, aim of this paper is to propose a novel algorithm for records selection to employ in risk analysis, which is able to control the dispersion without resorting to the artificial scaling of the accelerograms and thus can be easily employed in three-dimensional structural analysis. It is based on the idea to control the record-to-record variability rather than to reduce it by scaling the accelerograms. This intends to solve the old problem of selecting efficient intensity measure (IM), which characterized the last decade of research on Performance based earthquake engineering. In this respect a different use of UHS is here proposed, which has been implemented in a Matlab-based code named SCoReS.