Automated extraction of building features is a great aid in synthesizing building maps from radar data. In this paper, a model-based method is described to detect and classify canonical scatters, such as corners and planar walls, inside a building. Once corners and walls have been located, a building map can be synthesized. To detect and classify the canonical scatterers, sparse reconstruction with an overcomplete dictionary is used. The dictionary is tuned to phase models that are specific to the different canonical scatterers, i.e., a linear phase change for walls and a quadratic phase change for corners. This is a potentially robust method for detecting canonical scatterers because the polynomial degree of the measured phase is preserved, even after propagation through a wall. Building-feature extraction results obtained with actually stand-off measured through-wall radar data are discussed.Index Terms-Overcomplete dictionary (OCD), radar imaging, sparse reconstruction, through-wall radar.