The formation and demolition of vacant houses are the most visible sign of city shrinking and revitalization. Timely detection of vacant houses has become an inevitable task to aid the “Smart City” initiative. Two pressing problems exist for vacant houses, however: (1) No publicly accessible information is available at the individual house level and (2) the decennial census survey does not catch up with the rapidly changing status of vacant houses. To this end, remote sensing provides a low-cost avenue for detecting vacant houses. Traditionally, remote sensing was accredited for its success in deriving biophysical parameters of human settlements, such as the presence and physical size of buildings. It is still a challenge, though, to infer the functions of buildings, such as land-use types and occupancy status. In this study, we aim to detect individual vacant houses with very-high-resolution remote sensing images through a smart machine learning method. Our proposed method entails three steps: ground-truth data collection, classification, and feature selection. As a result, a new building change detection method was developed to collect ground-truth vacant house data from multitemporal images. Important features for classification of houses were identified. Subsequently, we carried out a classification of vacant houses and yielded promising results. Furthermore, the results indicate that both the area of the vacant house parcels and the healthy conditions of the surrounding vegetation contribute most to the detection accuracy. Our work shows the potential of using remote sensing to detect individual vacant houses at a large spatial extent. Key Words: machine learning, remote sensing, smart city, vacant house.