To improve the performance of the statistical approach to license plate recognition(LPR),an algorithm based on Gaussian mixture models(GMM) is proposed in this paper.First,the features of plate character are detected by using Gabor filter.Then the class conditional probability density is described with GMM.Then a Bayesian classifier is constructed according to the class conditional probability density.The GMM parameters can be estimated by expectation maximum(EM) algorithm.Using the algorithm,a high recognition rate can be achieved.Experimental results demonstrate the algorithm is feasible,robust and applicable.