A novel license plate localization algorithm is presented based on the saliency map. The feature maps with respect to intensity, color and orientation are firstly extracted from the input color RGB image by using the bottom-up visual attention model. Moreover, these feature maps are combined into a saliency map. After the binarization of saliency map, several plate candidates are obtained. Then, vertical edges, the aspect ratio, and hopping frequency of the brightness of the license plate region are used to eliminate the undesired candidates and determine the real license plate region. The comparison experimentations show that the proposed method can reduce the pseudo-plate number due to the background environment and have strong environmental adaptability, anti-noise ability as well as high localization rate.