The methods using spectral information alone are often ineffective due to the spectral similarity between roads and other artificial structures with impervious surface. This paper proposed a knowledge-based method for urban road extraction by including spatial texture information. The spatial texture feature was firstly extracted by the local Moran's I and the derived texture was added to the spectral bands of images for image segmentation. Then, features like brightness, standard deviation, rectangularity, aspect ratio and area were selected to form the hypothesis and verification model. Finally, roads were extracted by applying the models and were post-processed on the basis of mathematical morphology. This new method was evaluated by a 0.1m aerial image. The results show that the extraction accuracy reaches about 88% by using the proposed method, 5% higher than the corresponding images without the spatial texture information.