Correct and accurate road data is important for various applications such as urban planning, traffic control, vehicle navigation and safety. It is also well known that automated road extraction from aerial images can significantly reduce the cost of data acquisition and update. In this paper, a novel and effective approach that exploit both road geometrical and photometrical attributes to detect road network using aerial images is proposed. The algorithm starts by constructing DSM from stereo images, and then true orthophoto is generated by combining aerial images and DSM. As a result, the image displacement that presents in unrectified imagery is overcome. Then high quality feature and cues that consist of 3-D straight edges and road markings are extracted and combined to increase the success rate and the reliability of the generated results. Before image classification algorithm is applied, the original RGB image data is transformed into different color spaces to enhance object feature, and principal component transformation is employed to analyse the original image data and achieve a high data reduction ratio. Next, ISODATA clustering algorithm is used to separate road from other ground objects. After the detection of road feature, the missing road segments are bridged using the extracted cues and reasonable hypotheses. Finally, the generated road network is evaluated to test the performance of the proposed approach, in which the datasets provided by Main Road department of Queensland are used. The experiment result proves the effectiveness of our approach.