In this paper we present a location aided knowledge extraction routing (LAKER) protocol for MANETs, which utilizes a combination of caching strategy in dynamic source routing (DSR) and limited flooding area in location aided routing (LAR) protocol. The key novelty of LAKER is that it can gradually discover knowledge of topological characteristics such as population density distribution of the network. This knowledge can be organized in the form of a set of guiding/spl I.bar/routes, which includes a chain of important positions between a pair of source and destination locations. The guiding/spl I.bar/route information is learned during the route discovery phase, and it can be used to guide future route discovery process in a more efficient manner. LAKER is especially suitable for mobility models where nodes are not uniformly distributed. LAKER can exploit the topological characteristics in these models and limit the search space in route discovery process in a more refined granularity. Simulation results show that LAKER outperforms LAR and DSR in term of routing overhead, saving up to 30% broadcast routing message compared to the LAR approach.