Summary The modal combination rules commonly used in response spectrum analyses implicitly assume that the peak factor associated with the response quantity of interest is equal to the peak factors of the contributing modal responses. In this paper, we examine the validity of this assumption and demonstrate that it causes the modal combination rules to over‐represent the contribution of the higher modes of vibration to the total response and under‐represent the contribution of the lower modes. Consequently, a response‐spectrum‐based analysis can yield a biased estimate for the peak value of a response quantity when two or more well‐separated modal frequencies make significant contributions to the total response. To correct this potential bias in response‐spectrum‐based estimates, we develop a procedure for estimating the peak factors that is suitable to the response spectrum analysis calculations commonly used in the current design practice. Examples are presented to demonstrate the proper use and potential impact of the proposed procedure. Copyright © 2015 John Wiley & Sons, Ltd.