Summary | The response of a non‐circular structure strongly depends on the orientation of the horizontal ground motion components vis‐à‐vis the structure’s principal axes. At the same time, the structural response is also a function of accelerogram characteristics that give rise to considerable record‐to‐record variability even when the incident angle is neglected. Therefore, when the structural orientation relative to the fault geometry is unknown and we have limited resources for estimating the distribution of structural response given the seismic intensity, the question arises as to whether it is preferable to use (1) few records rotated to multiple orientations, (2) many records, each at a random incident angle, or (3) some combination of the two. To this purpose, we subjected several single‐degree‐of‐freedom systems and one plan‐asymmetric multi‐degree‐of‐freedom structure to a pulsive and a non‐pulsive set of ground motions using different combinations of record set size and incident angle rotations. In all cases, the natural record‐to‐record variability of the (unrotated) waveforms clearly outweighed the influence of the record orientation. In addition, the choice of an intensity measure that utilizes the geometric mean of spectral accelerations in both horizontal axes at one or more periods of vibration was found to further enhance this difference, essentially nullifying the already small effect of the incident angle. In all cases, spending any significant proportion of the limited number of dynamic analyses to incorporate the effect of incident angle was detrimental to the fidelity of the estimated performance.