Because the scene where license plate recognition system is located in is more and more complicated, license plate location method should have better performance. A new license plate location method based on Tophat transformation and text texture was introduced. First, Tophat transformation was applied to restrain background and eliminate variant illumination of image, and then binary operator and connected component analysis were applied. Second, the VPM(vertical projection map) of image was transferred to frequency space by DCT(discrete cosine transformation). The low frequency coefficients were used to reconstruct VPM. The reconstructed VPM was smoother, so region texture could be described essentially the wave crests and troughs number, statistic of reconstructed VPM. Last, pattern vectors were made up of low frequency component of frequency descriptors and statistic of reconstructed VPM, and were classified into license plate regions and non-license plate regions by SVM(support vector machine). Experimental results show that the method has good performance even when image has low quality and license plate is located in complicated natural scene.