License plate holds the very small proportion in the whole vehicle image, and the color, size and location for License plate is also uncertain. Fast and accurate localization for license plate is a difficult problem in the license plate recognition process. This paper presents a geometry license plate localization method based on machine vision. This method has very good compatibility to the gradation linear and nonlinear change, moreover support criterion and angle change. Experimental results indicate that the localization deviation does not surpass 1.0pixel and the angular deviation does not surpass 1.0 degrees under same level test condition. Robustness of this localization is basic consistent with Cognex Visionpro. The license plate localization can adapt the low signal-to-noise ratio, the low contrast gradient, the fuzzy situations and so on., which is also simple, practical and is better than traditional zone location and edge localization method.