Trust and reputation systems have proven to be essential to enforcing cooperative behavior in peer-to-peer networks. We briefly describe the current approaches to building reputation systems: social networks formation, probabilistic estimation and game theoretic models. We then observe that all of the current models make a number of simplifying assumptions that may not necessarily hold in real networks, such as either irrational (probabilistic) or completely rational behavior, instant propagation of reputation information and homogeneity of interactions. We argue that dropping those assumptions and allowing more degrees of freedom is necessary in order to construct more realistic and rich reputation models. We support our argument by citing reputation research done in economics, evolutionary psychology, biology and sociology and and consider models that take into account adaptive behavior changes, co-evolution of behaviors, bounded rationality and variable interaction patterns. We then outline how those complexities can be dealt with and point out main directions for the future study of more realistic and less constrained reputation models that can potentially lead to construction of more secure, responsive and cooperative peer-to-peer systems.