Calculation of structural collapse risk using non-linear response history analysis requires the selection of ground motions at different intensity levels. These selected ground motions should be consistent with the seismic hazard at the site under consideration. Source-specific, conditional distributions of ground motion duration and response spectra are proposed as targets to select hazard-consistent ground motions. Target distributions of duration are computed using a prediction equation for duration and earthquake source characteristics (e.g. source type, magnitude, and distance) obtained from seismic hazard deaggregation calculations, conditional on the exceedance of a spectral acceleration value corresponding to a specific hazard level. The correlation between the residuals (£ values) of response spectral ordinates and duration are accounted for in the calculation procedure. Sample calculations are performed for three sites in Western USA: Seattle (Washington), Eugene (Oregon), and San Francisco (California) to illustrate the contribution of interface earthquakes in subduction zones that are known to produce long duration ground motions. Previous studies have found that long duration ground motions, on average, predict lower collapse capacities than short duration ground motions such as the FEMA P695 far field records and moderate to large amplitude records from the PEER NGA West2 database, which are commonly used for collapse capacity estimation. The examples presented in this paper illustrate that the use of only short duration records for sites where interface earthquakes contribute significantly to the seismic hazard can lead to an over-estimation of collapse capacity and un-conservative structural designs.