Extraction of roads from the high-resolution SAR imagery is described. The method employs Hough Transform to identify roads, followed by a series of pre-processes. Gaussian probability iteration method instead of usually used CFAR detector or threshold segmentation is employed to identify the regions of interest. The generally used Hough Transform is replaced by the average Hough Transform. The peak values in the transform area are detected by a maximum search integrating global CFAR detector. Finally, a unique geometric constraint is used to discriminate the potential roads. High-resolution SAR images of MSTAR Redstone are used to illustrate our method, and the performance is satisfactory.