Abstract The coverage area of Access Points (APs) in Wireless Local Area Networks (WLANs) often overlaps considerably. Hence, a station can potentially associate with many APs. In traditional IEEE 802.11 systems, the station associates to the AP with the strongest signal. This strategy may result in load imbalance between APs and thus low overall network throughput. This paper proposes a new mechanism for selecting the “best” AP based on a novel available bandwidth estimation scheme. The available bandwidth provided by an AP depends mainly on the signal quality and the load on the wireless channel. Based on measurements we first analyze how those factors vary stochastically over time and motivate why a frequent estimation of available bandwidth is necessary. We then develop BEST-AP, a system for Bandwidth ESTimation of Access Points, which uses regular data traffic to estimate the available bandwidth from all APs in reach in a non-intrusive way, even if they are on a different channel. Based on OpenFlow, BEST-AP allows the station to be associated with multiple APs simultaneously and to switch between APs with low overhead. Using the available bandwidth estimates, the system exploits the “best” AP for longer duration while probing the less good APs for shorter durations to update the bandwidth estimations. The evaluation in a WLAN testbed shows that with background load created from real WLAN traces, the dynamic selection of APs improves the throughput of a station by around 81%, compared to a static selection. When the station is mobile, the throughput increases by 176% on average.