We propose a system architecture and algorithm for improved localization of indoor service robots. Many studies on sensor network based localization systems have been reported for indoor service robots. Previous sensor network based localization systems have used a sensor-agent which is composed of only one kind of sensor, but such an arrangement may have difficulty adapting to changing environmental conditions. Therefore, we configure the sensor-agent with different kinds of sensors using a sensor fusion concept. To do so, we propose a novel architecture for a sensor network based localization system. We further propose a localization algorithm which processes the position data from each sensor-agent. Finally, we apply the proposed localization algorithm to real indoor service robots and demonstrate the improved localization ability of the service robots.