This paper introduces an approach to architectural “distant reading”: the use of computational methods to analyze architectural data in order to derive spatial insights from—and explore new questions concerning—large collections of architectural work. Through a case study comprising a dataset of religious buildings, we show how we may use machine learning techniques to identify typological and functional traits from building plans. We find that spatial structure, rather than local features, is particularly effective in supporting this type of analysis. Further, we speculate on the potential of this computational method to enrich architectural design, research, and criticism by, for example, enabling new ways of thinking about architectural concepts such as typology in ways that reflect gradual variations, rather than sharp distinctions.