Most building design codes distinguish structural reliability levels in terms of failure consequences, for which they normally define consequence classes based on building type and use. Although readily applicable in everyday practice, that approach may entail adopting inconsistent safety requirements. Such a significant drawback could be minimised by establishing separate reliability levels for key members on the grounds of the potential consequences of their collapse. Further to those concerns, this paper proposes a series of consequence classes determined in keeping with the number of persons at risk in a given collapse scenario and the extent of the respective damage. Consequence class-related models for predicting loss of life are derived from statistical assessments of data on over 150 collapsed buildings. The models developed estimate the number of fatalities and conditional probability of death of building users under given collapse circumstances. In addition to their utility in establishing target reliability values, these models can be applied in risk analysis of specific building structures, especially where the potential consequences of failure are high.