In this thesis, we define and analyze communication scenarios in which the source and destination cooperate across noncoherent two-way fading channels. As in practical communication systems, we constrain the bandwidth and power resources available at both nodes. By constraining resources, we automatically produce situations in which neither the source nor the destination knows the fading state perfectly, and the nodes are unable to perfectly share knowledge with one another. Thus, not only is the channel state information imperfect, but the ability to cooperate in a coordinated fashion is also impaired. As a result, performance predictions based on perfect channel state information and perfect coordination can prove highly unrealistic. However, we have discovered that even in the presence of imperfect channel state information and imprecise coordination, cooperative communication over two-way channels offers significant performance gains over comparable one-way systems, and we have developed novel schemes to leverage cooperative gains in the presence of uncertainty. In the first part of this work, we analyze the reliability of a system that uses channel inversion power control to combat the effects of fading. We demonstrate that a simplistic approach to training leaves the system crippled by channel estimation errors. As an alternative, we propose a novel two-way training scheme that limits the impact of estimation error and provides the best known performance of any system subjected to full resource accounting. We then show that the benefits of two-way training extend to systems with multiple transmit antennas and two-way systems in which the data flow is bidirectional. In the second part, we employ a more general system model and study the precise amount of mutual information lost to uncertain channel state information at the destination. Our results illuminate the relationship between the loss of mutual information and the entropy of the channel. We also capture the effect of coherence time and the degree to which data-aided estimation can mitigate channel estimation errors. We then apply our results to quantify the performance of a number of practical systems and demonstrate the effectiveness of two-way cooperation in realistic scenarios.