U.S. national building codes refer to the ASCE/SEI‐7 provisions for selecting and scaling ground motions for use in nonlinear response history analysis of structures. Because the limiting values for the number of records in the ASCE/SEI‐7 are based on engineering experience, this study examines the required number of records statistically, such that the scaled records provide accurate, efficient, and consistent estimates of “true” structural responses. Based on elastic–perfectly plastic and bilinear single‐degree‐of‐freedom systems, the ASCE/SEI‐7 scaling procedure is applied to 480 sets of ground motions; the number of records in these sets varies from three to ten. As compared to benchmark responses, it is demonstrated that the ASCE/SEI‐7 scaling procedure is conservative if fewer than seven ground motions are employed. Utilizing seven or more randomly selected records provides more accurate estimate of the responses. Selecting records based on their spectral shape and design spectral acceleration increases the accuracy and efficiency of the procedure.

The California Building Code requires at least two ground motion components for the three‐dimensional (3‐D) response history analysis (RHA) of structures. For near‐fault sites, these records should be rotated to fault‐normal/fault‐parallel (FN/FP) directions, and two RHA analyses should be performed separately. This approach is assumed to lead to two sets of responses that envelope the range of possible responses over all non‐redundant rotation angles. This assumption is examined here using 3‐D computer models of single‐story structures having symmetric and asymmetric plans subjected to a suite of bidirectional earthquake ground motions. The influence that the rotation angle has on several engineering demand parameters is investigated in linear and nonlinear domains to evaluate the use of the FN/FP directions, and the maximum direction (MD). The statistical evaluation suggests that RHAs should be conducted by rotating a set of records to the MD and FN/FP directions, and taking the maximum response values from these analyses as design values.