A　widespread　approach　for　the　prediction　of　the　structural　response　as　function　of　the　ground　motion　intensity　is　based　on　the　Cloud　Analysis:　once　a　set　of　points　representing　the　engineering　demand　parameter　(EDP)　values　is　obtained　as　function　of　the　selected　seismic　intensity　measure　(IM)　for　a　collection　of　unscaled　earthquake　records,　a　regression　analysis　is　performed　by　assuming　a　specific　functional　form　to　correlate　these　variables.　Within　this　framework,　many　studies　have　been　devoted　so　far　to　evaluate　the　effectiveness　of　several　IMs　in　estimating　the　EDPs　through　intrinsically　linear　functional　forms,　but　it　is　still　unknown　to　what　extent　the　use　of　the　linear　regression　analysis　affects　the　quality　of　the　final　results.　This　paper　is　intended　to　provide　an　answer　to　such　question　by　means　of　the　calibration　of　suitable　nonlinear　combinations　of　scalar　IMs,　whose　statistical　performances　are　compared　with　those　obtained　by　using　the　functional　form　usually　adopted　for　linear　regression-based　calibrations.　Specifically,　the　Evolutionary　Polynomial　Regression　technique　is　adopted　to　calibrate　nonlinear　regression　models　for　the　prediction　of　maximum　inter-story　drift　ratio　and　maximum　floor　acceleration.　The　comparative　analysis　is　performed　for　fixed-base　and　base-isolated　reinforced　concrete　buildings　subjected　to　ordinary　or　pulse-like　ground　motion　taking　into　account　accuracy,　complexity,　efficiency　and　sufficiency.　Final　results　demonstrate　that　the　linear　regression　analysis　is　suitable　for　fixed-base　reinforced　concrete　buildings,　but　nonlinear　regression　models　provide　better　estimates.　On　the　other　hand,　the　linear　regression　analysis　can　introduce　a　significant　bias　in　the　seismic　response　prediction　of　base-isolated　buildings,　and　nonlinear　regression　models　are　deemed　more　appropriate.