According to the prior information of license plate, a new algorithm for characters segmentation of license plate based on variance projection and mean filter is proposed. Firstly, some preprocessing is performed toward the license plate images and image binarization is adopted after the processes. Then the horizontal boundaries and rivet are removed by using horizontal projection of license plate. Thus character separation of the license plate is implemented by using vertical projection which is processed based on mean filter, located the best segmentation points of adjacent characters. The experimental result shows that this algorithm has a good effect on the segmentation.