The authors present a vehicle license plate recognition algorithm based on improved pattern matching. This algorithm is made up of two parts plate characters segmentation and characters recognition. In the license plate characters segmentation algorithm, dynamic thresholding is utilized to process license plates, and coordinate marking is utilized to divide character. In the characters recognizable algorithm, improved pattern matching is applied to extract character characteristic storehouses, and characters recognition is realized by matched with line and row characteristic templates. The algorithm is implemented in VC++, and experiments prove that the rate of characters segmentation and recognition is high against stained car licenses.