In the present paper, a method is proposed that uses the geomagnetic field to estimate the positions of mobile robots in indoor environment. The reinforced concrete, H-beams, and power systems in buildings distort the geomagnetic field and these characteristics are used as regional features to estimate the positions of mobile robots. Another method is also proposed that prepares geomagnetic field maps using the model information obtained from the characteristics of sensors mounted on mobile robots and the position information to efficiently access the database. The extended Kalman filter is used to fuse robot motion models with geomagnetic field sensor models. The performance of the proposed algorithm is verified through Ground Truth experiments.