The problem of multi-robot simultaneous localization in an unknown environment based on relative observation was studied. Each robot was equipped with proprioceptive sensors and exteroceptive sensors. Exteroceptive sensors provided relative observations between robots, such as relative distance and bearing. Considering the nonlinear problem, we used extended Kalman filter to fuse proprioceptive and exteroceptive data. Different relative observations and the number of the robots were studied and simulated. The filter structure for each case was presented and the performances of the algorithm were compared as to the accuracy and other aspects. Simulation results prove that the location accuracy has been improved effectively by using relative observations among robots.