In this paper a method for online identification of satellite moment of inertia tensor parameters based on recursive least squares method, is presented. It is assumed that the satellite actuators are three orthogonal reaction wheels. Dynamic equations of the satellite are extracted in a special manner. The only available sensor is a three axes rate gyro which measures the angular velocity of satellite in the body coordinate system. Due to existence of noise in this sensor, the regressor matrix used in least squares method, changes stochastically. So in this case, the classic least squares method is not useful, and it cannot converge. For solving this problem, a modified least squares method with robust scheme is presented and its stability is proved using Lyapunov stability theory. The presented method can be used online in presence of measurement noise and other sensor imperfections. Simulation results have shown that this method can identify inertia parameters of the satellite with less than 3 percent error comparing to real parameters before and after changes.