Experimental and computational investigations addressing how various neural functions are achieved in the brain converged in recent years to a unified idea that the neural activity underlying most of the cognitive functions is distributed over large scale networks comprising various cortical and subcortical areas. Modeling approaches represent these areas and their connections using diverse models of neurocomputational units engaged in graph-like or neural field-like structures. Regardless of the manner of network implementation, simulations of large scale networks have encountered significant difficulties mainly due to the time delay introduced by the long range connections. To decrease the computational effort, it is common to assume severe approximations to simplify the descriptions of the neural dynamics associated with the system's units. In this dissertation we propose an alternative framework allowing the prevention of such strong assumptions while efficiently representing th e dynamics of a complex neural network. First, we consider the dynamics of small scale networks of globally coupled non-identical excitatory and inhibitory neurons, which could realistically instantiate a neurocomputational unit. We identify the most significant dynamical features the neural population exhibits in different parametric configuration, including multi-cluster dynamics, multi-scale synchronization and oscillator death. Then, using mode decomposition techniques, we construct analytically low dimensional representations of the network dynamics and show that these reduced systems capture the dynamical features of the entire neural population. The cases of linear and synaptic coupling are discussed in detail. In chapter 5, we extend this approach for spatially extended neural networks. / We consider the dynamical behavior of a neural field-like network, which incorporates many biologically realistic characteristics such as heterogeneous local and global connectivity as well as dispersion in the neural membrane excitability. We show that in this case as well, we can construct a reduced representation, which may capture well the dynamical features of the full system. The method outlined in this dissertation provides a consistent way to represent complex dynamical features of various neural networks in a computationally efficient manner. / by Roxana A. Stefanescu. / Thesis (Ph.D.)--Florida Atlantic University, 2009. / Includes bibliography. / Electronic reproduction. Boca Raton, Fla., 2009. Mode of access: World Wide Web.