Assessing seismic vulnerability at large scales requires accurate attribution of individual buildings to more general typological classes that are representative of the seismic behavior of the buildings sharing same attributes. One-by-one evaluation of all buildings is a time-and-money demanding process. Detailed individual evaluations are only suitable for strategic buildings, such as hospitals and other buildings with a central role in the emergency post-earthquake phase. For other buildings simplified approaches are needed. The definition of a taxonomy that contains the most widespread typological classes as well as performing the attribution of the appropriate class to each building are central issues for reliable seismic assessment at large scales. A fast, yet accurate, survey process is needed to attribute a correct class to each building composing the urban system. Even surveying buildings with the goal to determine classes is not as time demanding as detailed evaluations of each building, this process still requires large amounts of time and qualified personnel. However, nowadays several databases are available and provide useful information. In this paper, attributes that are available in such public databases are used to perform class attribution at large scales based on previous data-mining on a small subset of an entire city. The association-rule learning (ARL) is used to find links between building attributes and typological classes. Accuracy of wide spreading these links learned on less than 250 buildings of a specific district is evaluated in terms of class attribution and seismic vulnerability prediction. By considering only three attributes available on public databases (i.e. period of construction, number of floors, and shape of the roof) the time needed to provide seismic vulnerability scenarios at city scale is significantly reduced, while accuracy is reduced by less than 5%.