Urban area is a very typical non-uniform area with complex surface type and fast temperature change. It is a very challenging task to study the thermal anomaly detection and extraction of urban buildings on-orbit. A thermal anomaly detection and extraction method of urban buildings on-orbit using LANDSAT images is presented. Firstly, urban buildings are automatically extracted based on the multi-exponential feature and geometric feature, then the single-channel algorithm is utilized to retrieve the temperature of the buildings, and finally the on-orbit thermal anomaly detection and extraction of urban buildings is completed according to the temperature inversion results. The feasibility and effectiveness of the method are verified in Wuhan city referenced to ASTER officially surface temperature products. Experiment results show that the proposed method can accurately extract the urban buildings, and the root-mean-square error of target temperature inversion is less than 3 K, indicating that this method is an effective method for detecting and extracting thermal anomalies of urban buildings on-orbit.