Abstract | Earthquake loss estimation studies for portfolios of buildings are performed routinely for a variety of applications, ranging from risk evaluation and mitigation to post-event emergency assessment. Essential ingredients of such analyses are the vulnerability functions appropriate for different building classes representing the regional inventory. When abundant empirical data are absent, which is often the case, these vulnerability functions are obtained numerically via dynamic analyses informed by sets of ground motion records, usually selected without specific criteria. The implicit assumption is that the vulnerability function of two identical buildings located at different sites in the region would be identical. However, the structural response estimates even of identical buildings are sensitive to the characteristics of the earthquakes that control the hazard at each site in the region. Hence, strictly speaking, vulnerability functions should be both structure- and site-specific. This consideration is always neglected in portfolio loss assessment where identical vulnerability functions are used for buildings in the same class regardless of where they are located. Developing a set of different, site-specific vulnerability functions for like buildings in the same class is, however, impractical. To address this complexity, a record selection scheme is proposed that employs the conditional spectrum (CS) method but uses as a conditioning Intensity Measure (IM) the spectral acceleration (geometrically) averaged over a period range. This conditioning IM is more in sync with the response of a class of buildings rather than with the response of any single one. In addition, this method modifies the standard CS approach by incorporating within a single target CS the variation of the target spectra at multiple sites. An example of this method illustrates the development of a vulnerability function that is consistent with the regional hazard for a class of tall buildings.