Abstract: Odometric error modeling for mobile robot is a basis of localization. Most of the approaches to odometric error modeling are designed for some special driving-type robot up to now. And the unbounded odometric long term error, which degrades localization precision after long-distance movement, is not often able to be compensated in real-time. Therefore, a general approach to odometric error modeling for mobile robot is proposed with respect to both synchronous-drive roller robot and differential-drive roller robot. The method assumes that the robot path to be approximately to circular arcs. The approximate functions relationships between the process input of odometry and non-systematic errors as well as systematic errors are derived based on the odometric error propagation law. Further, a new localization algorithm for mobile robot is proposed, which is used to online compensate the accumulative errors of odometry in the process of localization. The experiments show that the new localization algorithm reduces the accumulative errors of odometry efficiently, and improves the localization precision remarkably.