This paper illustrates a probabilistic-based methodology for quantifying the collapse potential of structural systems, which can provide us with more accurate estimates of losses induced by earthquakes. Applications of this methodology for assessment of collapse potential of existing buildings and design for collapse safety are demonstrated by equations and example. The collapse potential is represented by the probability of collapse at discrete hazard levels and on an annualized basis (mean annual frequency). The basic ingredient of the proposed methodology is a ‘collapse fragility curve’ which expresses the probability of collapse as a function of the selected ground motion intensity measure. The process for estimating the collapse fragility using scalar and vector-valued ground motion intensity measure is demonstrated. The proposed assessment and design processes do incorporate the effect of aleatory and epistemic uncertainties. It was shown by example that the uncertainties, both aleatory and epistemic, have a significant effect on the outcome of the conceptual design for collapse safety. Copyright © 2007 John Wiley & Sons, Ltd.