Aiming at the problem of extracting thermal fault area in detecting power electronic using infrared imaging, a method of extracting thermal fault area is proposed, which is based on the Mediodshift with local clustering. Firstly, the clustering center is initialized according to the gray level characteristics of thermal faults. Then, the neighborhood pixels are clustered by combining Mediodshift clustering method. In order to promote the performance in clustering of pixels as far as possible, the adjustment strategy for neighborhood gray level is introduced. At the same time, in order to improve the clustering efficiency, a clustering thresholding segmentation mechanism from high to low is adopted, which enables Mediodshift algorithm to cluster the pixels of fault region quickly and to extract the thermal fault region effectively from infrared image. Finally, Experiment on some classic electronic fault images show that the region extraction performance of our method has its ability to extract the region, and is better than the performance of some existing methods.