GPS has been commonly used for vehicle localization. However, since GPS signals are very sensitive to terrain and interference, they may become unavailable in places such as lower layers of multilayer bridges, streets besides high buildings, tunnels, etc. To address this issue, various techniques have been proposed. Nevertheless, none of them are applicable to safety-critical and highly mobile vehicular network applications, which require high localization accuracy anywhere and anytime. Meanwhile, with the advent of radio frequency identification (RFID) systems on roads (RSR), passive RFID tags with road-related information are expected to be deployed on road surfaces for driving safety and efficiency. In this paper, we introduce a novel RFID-based approach for vehicle localization in GPS-less road environments, where passive RFID tags are deployed on the roads and readers are installed on vehicles. To achieve accurate vehicle localization, we propose an error-cognitive localization system, which leverages both readerto-tag and vehicle-to-vehicle communications to adaptively cognize and correct localization errors. Extensive simulation and experiment have verified the effectiveness of our proposed system. Particularly, our field tests show that, with sparsely deployed tags, which cost just a few dollars per kilometer, a vehicle can localize itself with submeter errors on average.