Dynamics of link bandwidth of a wireless link, which changes frequently and abruptly due to the dynamic channel sharing, fading, and mobility, is of interest to adaptive network applications and communication protocols. This paper presents a novel approach to estimate wireless link bandwidth based on radio signal-to-noise ratio (SNR). Unlike traditional methods that send probe packets, our method is non-intrusive to the wireless network since in IEEE 802.11 wireless local area networks, SNR information is provided by the physical layer for the MAC- and upper layers’ functionality. Theoretical analysis and experimental observation indicate a nonlinear relationship between SNR and the wireless bandwidth. Based on this, nonlinear models using neural network and Bayesian inference methods are proposed and evaluated on data collected in 802.11b wireless networks. The effectiveness of our method under various environments and scenarios has been studied.