This research examines the configuration and coordination of cognitive radio networks and seeks whether it is possible to efficiently configure a cognitive radio network using control algorithms that can independently manage and coordinate connections among radio nodes. Such control algorithms have to manage a cognitive radio in two ways - first, they must internally configure the radio to appropriately fulfill the users demand without causing interference to outside users; and second, they must configure the radio to allow satisfactory communication with external nodes since the cognitive radios objective is to exchange data. To address this coordination and configuration problem, this thesis examines two distinct problems. The first problem relates to improving the choice of internal configuration parameters and involves using fractional factorial designs to run a series of line experiments to determine a predictive model for performance among a set of nodes for a specific set of inputs. Through simulation and implementation, this thesis demonstrates that the fractional factorial design can be used to accelerate the convergence of a variety of optimization techniques (e.g., genetic algorithms and gradient ascent techniques). The second problem relates to the issue of channel assignment for cognitive radio networks, wherein different nodes must agree on which channels they will use to communicate. Through a series of theoretical analysis, software simulation and hardware implementation, this work demonstrates that biologically-inspired local control algorithms are a feasible and worthwhile avenue for cognitive radio coordination and shows promising prospects for other areas in wireless systems such as sensor and ad hoc networks. This thesis demonstrates that local control based on biologically-inspired algorithms is well suited to the coordination of cognitive radio nodes and that this approach can cope well with limited sensing capabilities and only partial knowledge about the system. The algorithm however showed certain limitations when large network sizes came into play due to its probabilistic inner workings, however there exist additional components that could address these issues and warrant further investigation.