In many major metropolitan areas, taxi services have played an important role as a semipublic transportation mode without public support. However, there has not been much modeling effort - despite the importance of taxis in urban transportation systems - mainly because of the difficulty in modeling taxi drivers' behavior. This study models a taxi service system in urban areas, taking into account taxi drivers' knowledge of the transportation network built from their day-to-day experience. Passenger-seeking behavior by taxi drivers is modeled on the basis of their expected travel time and expected waiting time. The model considers the stochastic and dynamic transportation network and various levels of network knowledge on the part of drivers. This modeling approach provides flexibility in modeling the characteristics of taxi operation as well as understanding how taxi drivers' capability evolves. The study analyzes the fleet size of taxi service systems and the effects of the taxi company's information systems by considering quality and operational efficiency of taxi services, from both the passengers' and taxi operators' points of view. A simulation experiment shows that the taxi information system can provide benefits equivalent to increasing the number of taxis by 20% in regard to the quality of taxi service.