Abstract | Fragility curves are an important ingredient in the seismic loss assessment process. For a regional scale loss estimation, to reduce to reasonable levels the computational burden associated with determining the seismic demands for individual buildings, analytical seismic fragilities are instead evaluated on a broad building class basis. The latter process essentially involves representing a population of buildings having similar characteristics with a set of characteristic “index” buildings to avoid analyzing every single building within this population. For the definition and modeling of index buildings, two main options are currently available, these being (a) defining a limited number of index buildings to represent the class and modeling them with relatively complex, yet more accurate, MDOF systems, and (b) defining numerous index buildings to represent the class and modeling them with simplified approximate SDOF systems. Apparently, the dilemma of defining the optimal way to sample the index buildings comes down to the use of few MDOFs or many SDOFs. Despite the fact that the use of many SDOFs is a rather attractive option, given that they are an easy and computationally inexpensive choice in terms of both modeling and analysis, they are often a bad approximation of the actual problem. This is the case, for example, of tall or irregular buildings, where non-negligible higher modes render the SDOF approximation ineffective. Then, the more expensive and accurate MDOF option has to be employed. However, using a limited number of MDOFs to represent the class of interest inherently offers very little flexibility towards capturing individual buildings that might belong to that class yet their salient features do not necessarily match those of the “average” index building. Aggregating the results of all index buildings into a single class fragility means that one cannot provide a more accurate answer than the mean class fragility plus some dispersion, even if the building in question actually closely matches one of the underlying index ones. This may not matter for estimating long-term average losses over a region, but it becomes increasingly important as the size of the portfolio is reduced and individual structures stand out. To resolve the aforementioned issue, we propose here a method for adding substance back to the class fragility and consequently obtaining fine-grained attribute-driven fragility estimates. The term attribute-driven is key in our approach, since it implies that the process explicitly accounts for the specific characteristics of the building in question. It is essentially a meso-scale approach that stands between the building-specific FEMA P-58 style approach (micro-scale) and the building-class approach (macro-scale). Our testbed is a population of modern high-rise reinforced concrete buildings, represented by seven index buildings, for which we have evaluated fragility functions. With this information at hand, our proposed approach employs statistical methodologies for effectively disaggregating the index building fragility functions, to provide attribute-aware response and collapse fragility spot estimates for individual sample buildings, other than the index ones, that belong to the same class.