In order to overcome the drawbacks for gaining and matching the features in traditional automatic license plate recognition algorithms, this paper put forward an approach for license plate recognition with the scaleinvariant feature transform (SIFT) algorithm. In this approach, it will build the license sample template and obtain the features vector of license sample template and license image, including license difference of gauss(DoG) scalespace extreme detection, license key point localization, license orientation assignment and license SIFT features vector descriptor. And then it will realize the license plate recognition for matching vector with best bin first (BBF) algorithm. Afterwards, the approach is used for the character recognition of the plates. The experimental results show that this algorithm has very good success rates.