Automatic road extraction from aerial images can be very helpful in traffic control and vehicle guidance systems. Most of the road detection approaches are based on image segmentation algorithms. Color-based segmentation is very sensitive to light changes and consequently the change of weather condition affects the recognition rate of road detection systems. In order to reduce the light change effects, a novel Otsu thresholding-based approach for automatic road detection is proposed. Otsu method calculates the optimum threshold value, which efficiently separates the two classes of road parts and non-road parts, so that their inter-class variance is maximal. Conducted experiments confirm the performance of the proposed method for detection of roads in different lighting conditions whilst the experimental images are taken from different geographical areas.