This letter proposes a new synthetic aperture radar (SAR) image building detection method based on the bi-dimensional empirical mode decomposition (BEMD) algorithm, well adapted to nonlinear and nonstationary signals. First, the SAR image is decomposed by the BEMD algorithm to generate intrinsic mode functions (IMFs), and the IMFs are combined to extract bright regions and dark regions from the SAR image. Next, the Markov random field (MRF) model is used to cluster the SAR image, and according to the centroid of the dark region, the dark category with a complete edge is located. Finally, the building is detected by combining the bright region, the dark region, and the given shadow direction. Analytical and experimental evidence show that the proposed method has high detection accuracy and has wide applicability for medium, large, and irregular shaped buildings with shadows.