Characters Segmentation is one important step in the Automatic License Plate Recognition (ALPR) system. This paper proposed an improved method for one-row & two-row of Vietnam license plates (LP) characters segmentation. The proposed method consists of two main modules: the first task is pre-processing (image quantization, normalized image, adjust horizontal contours, morphology opening to remove noises), the second task is characters segmentation based on the method named peak-to-valley in order to segment the pictures in digit images getting the two bounds of the each digit segment according to the statistical parameter. We tested for 600 Vietnam LP (300 one-row LP, 300 two-row LP), the average rate of accuracy of our method is 98.03%.