Determining the location of a wireless client is a key problem for location-aware systems and for security applications. Many recent studies have used Bayesian methods to determine location from wireless LAN signals, but such methods have the drawback that a model must first be built from training data. The introduction of model error can drastically reduce the robustness of the location estimates, rendering most models incapable of tolerating hardware differences, channel variations, and intentional interferences from malicious users. This thesis describes the design, implementation and analysis of Golden-Eye, a robust wireless LAN location-sensing system that uses new techniques to address this problem. By fitting training data into Gaussian distributions and using relative signal strength, Golden-Eye works independent of the client's 802.11 implementation or transmission power level, making it suitable even for tracking clients that might be trying to hide their locations.