Efficient network management can improve the network performance and reduce the cost of system maintenance and administration. One of the important duties assigned to the network management system in a infrastructure wireless LAN (or simply wireless LAN) is to assign users to appropriate accessible access points (APs). The current AP selection schemes in wireless LANs cause an unbalanced load, which in turn reduces the performance of individual users. This has motivated intensive studies attempting to determine efficient methods to balance loads among different APs. Existing works either provide heuristic solutions without performance guarantees or provide centralized solutions which are not desirable for the AP selection problem. In this paper, we study the localized solutions that can provide performance guarantees. We model the AP selection problem as a matching problem in bipartite graph. Our objective is to maximize total load among all APs. We propose a class of localized heuristics based on different user knowledge models. For some of these localized heuristics, we prove that there exists a constant approximation ratio in terms of expected total load through mathematical analysis. Simulations are conducted to verify our results. Copyright © 2009 John Wiley & Sons, Ltd.