Abstract | Recent research has shown that structures may experience quite different response when the components of any given ground motion record are rotated in the horizontal plane. On the other hand, significant differences are also associated with using different ground motion records. This paper examines whether it is preferable to use a small set of records, each rotated to multiple incidence angles, or a larger set with fewer (or no) rotations, in the case we wish to perform a fixed number of nonlinear dynamic analyses and we have little site information. To this purpose we collected a relatively large, non-pulsive set of records and applied it to an elastoplastic single-degree-of-freedom (SDOF) system. Finally, bootstrapping was performed to investigate the effect of the number of rotation angles studied versus the number of records on the statistics of response. The results indicate that, in all cases, and especially when a small number of nonlinear analyses is allocated, the effect of record-torecord variability clearly outweighs the incidence angle influence. Using a small set of records can lead to unreliable results by both inaccurate estimation of the central value and severe underestimation of the dispersion.