The invention discloses a method of detecting a building in an SAR (Synthetic Aperture Radar) image based on an evidence theory algorithm. The method comprises realization steps: (1) an image is inputted; (2) an inputted SAR image with a to-be-detected building is filtered; (3) a saliency image is acquired; (4) 500 attention focuses are acquired; (5) the image after filtering is segmented; (6) the credibility of each connected area is calculated; (7) whether the credibility of each connected area is larger than 0.5 is judged, if yes, the step (9) is executed, or otherwise, the step (8) is executed; (8) the connected area is deleted; and (9) a final detection result is outputted. The credibility of each connected area is judged, the accuracy problem due to over reliance on an angular line detection result is overcome, the detection precision is improved, and the method can be applied to detection and recognition on the building in a high-resolution SAR image.