This paper presents a new license plate detection method based on the Canny edge detection and Sauvola binarization combine with Connected Components Analysis (CCA) and edge dilation. The proposed method was tested with 944 images in many different weather conditions and many different sizes of plates with only one configuration for all. We also build a ground-truth for each images and base on this to measure O, E factor. If the O, E factor satisfied the condition we will have a right result and vice verse. The right result in process was 921 over 944 images (97.56%).