This paper presents a precise localization method for autonomous driving systems by correcting the GPS bias error. Since GPS errors have systematic noise properties that change slowly with time, a stand-alone GPS cannot be used for localization of an autonomous vehicle. To compensate for this systematic bias error, several types of additional sources of information, including on-board motion sensors, camera vision systems, and a road map database, are applied to the localization system. The localization algorithm is based on a particle filter, because the measurement model related to the representation of the road geometry is described by a highly nonlinear function. The proposed localization algorithm was tested and verified through an autonomous driving test.