The user equilibrium concept in traffic assignment is based on fundamental assumptions: perfect information, rationality, and homogeneity. This study examines changes in the traffic pattern and the network behavior when these assumptions are relaxed. In order to relax these assumptions, we employ a day-to-day evolution approach and develop agent-based simulation models that include drivers’ learning model, preference model and preference sensitivity. Using the developed models, we investigate how each assumption affects network traffic patterns. The test results show that the assumption of perfect information is most influencing on the traffic assignment results. In addition, the prior information greatly affects on the route choice when drivers do not have perfect knowledge on network travel time.