Currently, synthetic aperture radar (SAR) is a hot research topic in the field ofMicrowaveRemote Sensing.Itpossessesmany incomparable advantages such as the capability towork at all tmi e and under allweather, high spatial resolution and strong penetrability through the ground surface and so on. Therefore, it is significant to extract information from SAR mi agery. It can not only compensate the deficiency in optical mi agery, but also help constructing the spatial database.Road information is a kind ofmost mi portant spatial basic information, which has a lot of significant applications,includingmilitary defense,mapmatching, database updating etc. Ratio of average(RoA) algorithm is a classicalmethod for road extraction from SAR mi agery developed in recent 20 years, however, because of SAR speckle noise and other factors, it stillhas some disadvantages. For example: low locality accuracy, thickened edges, higher false alarming rate and non-continuous road segments.Thispaper ismainlydivided into two sections to resolve these problems. In the firstsection,it introduces a local ratio detector for road extraction. But ithasdefectson the accuracy ofposition andwidening roadwidth. In order to tacle these problems,a pruningwindow is added into the process, which is used to remove effects of rivers and trees. Therefore, a thinned andmore accurate road map can be gained. Based on the above method , in the second section , we firstly describe the road linkingmethod using chain codes, define the chain code energy network and mi port the priorknowledge into processing course. Then, we analyze linking probability of line element in smi ulated road map and propose an approach to optmi ize the chain codes. Because the process is programmedwith classed computation, and need no search for all line segments so that it mi proves efficiency highly. At last, thismethod is applied to extracting the road network from realSAR mi age and its validity has been proved.