This paper is the first analytical work to exhibit the substantial gains resulting from applying site specific knowledge to frequency allocation in wireless networks. Two new site-specific knowledge-based frequency allocation algorithms are shown to outperform all other published work. Site specific knowledge refers to knowledge of building layouts, the locations and electrical properties of APs, users, and physical objects. We assume that a central network controller communicates with all APs, and has site specific knowledge which enables the controller to predict, a priori, the received power from any transmitter to any receiver. Optimal frequency assignments are based on predicted powers to minimize interference and maximize throughput. Our algorithms consistently yield high throughput gains irrespective of network topology, AP activity level, and the number of APs, rogue interferers, and available channels. Our algorithms outperform the best published algorithm by up to 3.68%, 8.95%, 13.6%, 15.1%, 25.8%, and 84.9% for 50, 25, 20, 15, 10, and 5 percentiles of user throughputs, respectively.