Summary | The selection of a scalar Intensity Measure (IM) for performing analytical vulnerability (loss) assessment across a building class is addressed. We investigate the ability of several IM choices to downgrade the effect of seismological parameters (sufficiency) as well as reduce the record‐to‐record variability (efficiency) for both highrise and lowrise sets of ‘index’ buildings. These characteristics are explored in unprecedented detail, employing comparisons and statistical significance testing at given levels of local engineering demand parameters (story drift ratios and peak floor accelerations) that relate to losses, instead of global variables such as the maximum interstory drift. Thus, a detailed limit‐state‐specific view is offered for the suitability of different scalar IMs for loss assessment. As expected, typical single‐period spectral values are found to introduce unwanted bias at high levels of scaling, both for a single as well as a class of buildings. On the other hand, the geometric mean of the spectral acceleration values estimated at several periods between the class‐average second‐mode and an elongated class‐average first‐mode period offers a practical choice that significantly reduces the spectral‐shape bias without requiring the development of new ground motion prediction equations. Given that record selection remains a site‐ and building‐specific process, such an improved IM can help achieve reliable estimates for building portfolios, as well as single structures, at no additional cost. Copyright © 2015 John Wiley & Sons, Ltd.