In any modern society, individuals interact to form opinions on various topics, including economic, political, and social aspects. Opinions evolve as the result of the continuous exchange of information among individuals and of the assimilation of information distributed by media. The impact of individuals' opinions on each other forms a network, and as the time progresses, their opinions change as a function of structure of such network. It is a central question whether this interaction and assimilation process leads to a socially benecial aggregation of information. My work mainly addresses complex problems in the analysis of opinion evolution in (i) heterogeneous societies and (ii) societies with large population under the influence of exogenous events. In both topics, bounded condence models of opinion dynamics are considered: an individual changes its opinion in a discrete fashion, by taking into consideration only those other agents whose beliefs are within a certain bound of condence from its own opinion. In the study of opinion dynamics in heterogeneous networks: agents are classied based on their interaction network, the existence of a leader group is established, and sufficient condition for the convergence of agent's opinions to a nal set of decisions is derived. In the study of opinion evolution in a large population driven by peer-to-peer interactions and exogenous events: the convergence of population's distribution to clusters concentrated around separate opinions is proved; an empirical upper bound on the largest population that a xed input can attract to its center is computed; a linear relation between this largest attracted population and systems parameters such as agents' bound of trust is conjectured; and performance of dierent manipulation strategies are evaluated based on the limited attraction range of the manipulator agent.