This project develops artificial intelligence and spatial modeling methods for the extraction of architectural and geospatial data from historical records. In particular, the project automates the process of extracting data from maps of over 10,000 American municipalities that were made by The Sanborn Map Company during the nineteenth and twentieth centuries for fire insurance companies to assess potential liability. These historical maps have value for urban and regional planning, but the use of these maps has previously been hindered by the lack of tools for efficiently extracting the data. Also, the methods developed in this project facilitate the creation of longitudinal datasets that permit further scientific research on the development of cities in the United States. The project also provides training opportunities for students and researchers. Expanding on previous methods, this project develops refined techniques for extracting data from the Sanborn maps, including building footprints, construction materials, building use (e.g., residential, commercial, industrial), and the numbers of stories. A key innovation is the advancement of methods for inferring the three-dimensional architecture and structure of buildings. This approach requires extracting additional building geometry and orientation from the Sanborn maps and ancillary high resolution orthoimagery data for extant buildings, then converting this information into graph representations for analytical processi